Datasets:

pisterlabs

/

promptset

like 0

2024-01-10

SBrandeis/transformers

src~transformers~models~openai~modeling_tf_openai.py

# coding=utf-8 # Copyright 2018 The OpenAI Team Authors and HuggingFace Inc. team. # Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ TF 2.0 OpenAI GPT model.""" from dataclasses import dataclass from typing import Optional, Tuple import tensorflow as tf from ...activations_tf import get_tf_activation from ...file_utils import ( ModelOutput, add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, replace_return_docstrings, ) from ...modeling_tf_outputs import TFBaseModelOutput, TFCausalLMOutput, TFSequenceClassifierOutput from ...modeling_tf_utils import ( TFCausalLanguageModelingLoss, TFConv1D, TFPreTrainedModel, TFSequenceClassificationLoss, TFSequenceSummary, TFSharedEmbeddings, get_initializer, input_processing, keras_serializable, shape_list, ) from ...utils import logging from .configuration_openai import OpenAIGPTConfig logger = logging.get_logger(__name__) _CONFIG_FOR_DOC = "OpenAIGPTConfig" _TOKENIZER_FOR_DOC = "OpenAIGPTTokenizer" TF_OPENAI_GPT_PRETRAINED_MODEL_ARCHIVE_LIST = [ "openai-gpt", # See all OpenAI GPT models at https://huggingface.co/models?filter=openai-gpt ] class TFAttention(tf.keras.layers.Layer): def __init__(self, nx, n_ctx, config, scale=False, **kwargs): super().__init__(**kwargs) n_state = nx # in Attention: n_state=768 (nx=n_embd) # [switch nx => n_state from Block to Attention to keep identical to TF implem] assert ( n_state % config.n_head == 0 ), f"Hidden dimension {n_state} not dividable by number of heads {config.n_head}" self.n_ctx = n_ctx self.n_head = config.n_head self.split_size = n_state self.scale = scale self.output_attentions = config.output_attentions self.c_attn = TFConv1D(n_state * 3, nx, initializer_range=config.initializer_range, name="c_attn") self.c_proj = TFConv1D(n_state, nx, initializer_range=config.initializer_range, name="c_proj") self.attn_dropout = tf.keras.layers.Dropout(config.attn_pdrop) self.resid_dropout = tf.keras.layers.Dropout(config.resid_pdrop) self.pruned_heads = set() def prune_heads(self, heads): pass @staticmethod def causal_attention_mask(nd, ns, dtype): """ 1's in the lower triangle, counting from the lower right corner. Same as tf.matrix_band_part(tf.ones([nd, ns]), -1, ns-nd), but doesn't produce garbage on TPUs. """ i = tf.range(nd)[:, None] j = tf.range(ns) m = i >= j - ns + nd return tf.cast(m, dtype) def _attn(self, q, k, v, attention_mask, head_mask, output_attentions, training=False): # q, k, v have shape [batch, heads, sequence, features] w = tf.matmul(q, k, transpose_b=True) if self.scale: dk = tf.cast(shape_list(k)[-1], tf.float32) # scale attention_scores w = w / tf.math.sqrt(dk) # w has shape [batch, heads, dst_sequence, src_sequence], where information flows from src to dst. _, _, nd, ns = shape_list(w) b = self.causal_attention_mask(nd, ns, dtype=w.dtype) b = tf.reshape(b, [1, 1, nd, ns]) w = w * b - 1e4 * (1 - b) if attention_mask is not None: # Apply the attention mask w = w + attention_mask w = tf.nn.softmax(w, axis=-1) w = self.attn_dropout(w, training=training) # Mask heads if we want to if head_mask is not None: w = w * head_mask outputs = [tf.matmul(w, v)] if output_attentions: outputs.append(w) return outputs def merge_heads(self, x): x = tf.transpose(x, [0, 2, 1, 3]) x_shape = shape_list(x) new_x_shape = x_shape[:-2] + [x_shape[-2] * x_shape[-1]] return tf.reshape(x, new_x_shape) def split_heads(self, x): x_shape = shape_list(x) new_x_shape = x_shape[:-1] + [self.n_head, x_shape[-1] // self.n_head] x = tf.reshape(x, new_x_shape) return tf.transpose(x, (0, 2, 1, 3)) # (batch, head, seq_length, head_features) def call(self, x, attention_mask, head_mask, output_attentions, training=False): x = self.c_attn(x) query, key, value = tf.split(x, 3, axis=2) query = self.split_heads(query) key = self.split_heads(key) value = self.split_heads(value) attn_outputs = self._attn(query, key, value, attention_mask, head_mask, output_attentions, training=training) a = attn_outputs[0] a = self.merge_heads(a) a = self.c_proj(a) a = self.resid_dropout(a, training=training) outputs = [a] + attn_outputs[1:] return outputs # a, (attentions) class TFMLP(tf.keras.layers.Layer): def __init__(self, n_state, config, **kwargs): super().__init__(**kwargs) nx = config.n_embd self.c_fc = TFConv1D(n_state, nx, initializer_range=config.initializer_range, name="c_fc") self.c_proj = TFConv1D(nx, n_state, initializer_range=config.initializer_range, name="c_proj") self.act = get_tf_activation("gelu") self.dropout = tf.keras.layers.Dropout(config.resid_pdrop) def call(self, x, training=False): h = self.act(self.c_fc(x)) h2 = self.c_proj(h) h2 = self.dropout(h2, training=training) return h2 class TFBlock(tf.keras.layers.Layer): def __init__(self, n_ctx, config, scale=False, **kwargs): super().__init__(**kwargs) nx = config.n_embd self.attn = TFAttention(nx, n_ctx, config, scale, name="attn") self.ln_1 = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_epsilon, name="ln_1") self.mlp = TFMLP(4 * nx, config, name="mlp") self.ln_2 = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_epsilon, name="ln_2") def call(self, x, attention_mask, head_mask, output_attentions, training=False): output_attn = self.attn(x, attention_mask, head_mask, output_attentions, training=training) a = output_attn[0] # output_attn: a, (attentions) n = self.ln_1(x + a) m = self.mlp(n, training=training) h = self.ln_2(n + m) outputs = [h] + output_attn[1:] return outputs # x, (attentions) @keras_serializable class TFOpenAIGPTMainLayer(tf.keras.layers.Layer): config_class = OpenAIGPTConfig def __init__(self, config, *inputs, **kwargs): super().__init__(*inputs, **kwargs) self.config = config self.output_hidden_states = config.output_hidden_states self.output_attentions = config.output_attentions self.return_dict = config.use_return_dict self.num_hidden_layers = config.n_layer self.vocab_size = config.vocab_size self.n_embd = config.n_embd self.tokens_embed = TFSharedEmbeddings( config.vocab_size, config.n_embd, initializer_range=config.initializer_range, name="tokens_embed" ) self.positions_embed = tf.keras.layers.Embedding( config.n_positions, config.n_embd, embeddings_initializer=get_initializer(config.initializer_range), name="positions_embed", ) self.drop = tf.keras.layers.Dropout(config.embd_pdrop) self.h = [TFBlock(config.n_ctx, config, scale=True, name="h_._{}".format(i)) for i in range(config.n_layer)] def get_input_embeddings(self): return self.tokens_embed def set_input_embeddings(self, value): self.tokens_embed.weight = value self.tokens_embed.vocab_size = shape_list(value)[0] def _prune_heads(self, heads_to_prune): """ Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer} """ raise NotImplementedError def call( self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, training=False, **kwargs, ): inputs = input_processing( func=self.call, config=self.config, input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, training=training, kwargs_call=kwargs, ) if inputs["input_ids"] is not None and inputs["inputs_embeds"] is not None: raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time") elif inputs["input_ids"] is not None: input_shape = shape_list(inputs["input_ids"]) inputs["input_ids"] = tf.reshape(inputs["input_ids"], [-1, input_shape[-1]]) elif inputs["inputs_embeds"] is not None: input_shape = shape_list(inputs["inputs_embeds"])[:-1] else: raise ValueError("You have to specify either input_ids or inputs_embeds") if inputs["position_ids"] is None: inputs["position_ids"] = tf.range(input_shape[-1], dtype=tf.int32)[tf.newaxis, :] if inputs["attention_mask"] is not None: # We create a 3D attention mask from a 2D tensor mask. # Sizes are [batch_size, 1, 1, to_seq_length] # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length] # this attention mask is more simple than the triangular masking of causal attention # used in OpenAI GPT, we just need to prepare the broadcast dimension here. inputs["attention_mask"] = inputs["attention_mask"][:, tf.newaxis, tf.newaxis, :] # Since attention_mask is 1.0 for positions we want to attend and 0.0 for # masked positions, this operation will create a tensor which is 0.0 for # positions we want to attend and -10000.0 for masked positions. # Since we are adding it to the raw scores before the softmax, this is # effectively the same as removing these entirely. inputs["attention_mask"] = tf.cast(inputs["attention_mask"], tf.float32) inputs["attention_mask"] = (1.0 - inputs["attention_mask"]) * -10000.0 else: inputs["attention_mask"] = None # Prepare head mask if needed # 1.0 in head_mask indicate we keep the head # attention_probs has shape bsz x n_heads x N x N # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads] # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length] if inputs["head_mask"] is not None: raise NotImplementedError else: inputs["head_mask"] = [None] * self.num_hidden_layers # head_mask = tf.constant([0] * self.num_hidden_layers) inputs["position_ids"] = tf.reshape(inputs["position_ids"], [-1, shape_list(inputs["position_ids"])[-1]]) if inputs["inputs_embeds"] is None: inputs["inputs_embeds"] = self.tokens_embed(inputs["input_ids"], mode="embedding") position_embeds = self.positions_embed(inputs["position_ids"]) if inputs["token_type_ids"] is not None: inputs["token_type_ids"] = tf.reshape( inputs["token_type_ids"], [-1, shape_list(inputs["token_type_ids"])[-1]] ) token_type_embeds = self.tokens_embed(inputs["token_type_ids"], mode="embedding") else: token_type_embeds = 0 hidden_states = inputs["inputs_embeds"] + position_embeds + token_type_embeds hidden_states = self.drop(hidden_states, training=inputs["training"]) output_shape = input_shape + [shape_list(hidden_states)[-1]] all_attentions = () if inputs["output_attentions"] else None all_hidden_states = () if inputs["output_hidden_states"] else None for i, block in enumerate(self.h): if inputs["output_hidden_states"]: all_hidden_states = all_hidden_states + (tf.reshape(hidden_states, output_shape),) outputs = block( hidden_states, inputs["attention_mask"], inputs["head_mask"][i], inputs["output_attentions"], training=inputs["training"], ) hidden_states = outputs[0] if inputs["output_attentions"]: all_attentions = all_attentions + (outputs[1],) hidden_states = tf.reshape(hidden_states, output_shape) # Add last hidden state if inputs["output_hidden_states"]: all_hidden_states = all_hidden_states + (hidden_states,) if inputs["output_attentions"]: # let the number of heads free (-1) so we can extract attention even after head pruning attention_output_shape = input_shape[:-1] + [-1] + shape_list(all_attentions[0])[-2:] all_attentions = tuple(tf.reshape(t, attention_output_shape) for t in all_attentions) if not inputs["return_dict"]: return tuple(v for v in [hidden_states, all_hidden_states, all_attentions] if v is not None) return TFBaseModelOutput( last_hidden_state=hidden_states, hidden_states=all_hidden_states, attentions=all_attentions, ) class TFOpenAIGPTPreTrainedModel(TFPreTrainedModel): """ An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. """ config_class = OpenAIGPTConfig base_model_prefix = "transformer" @tf.function( input_signature=[ { "input_ids": tf.TensorSpec((None, None), tf.int32, name="input_ids"), "attention_mask": tf.TensorSpec((None, None), tf.int32, name="attention_mask"), } ] ) def serving(self, inputs): output = self.call(inputs) return self.serving_output(output) @dataclass class TFOpenAIGPTDoubleHeadsModelOutput(ModelOutput): """ Base class for outputs of models predicting if two sentences are consecutive or not. Args: logits (:obj:`tf.Tensor` of shape :obj:`(batch_size, num_choices, sequence_length, config.vocab_size)`): Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). mc_logits (:obj:`tf.Tensor` of shape :obj:`(batch_size, num_choices)`): Prediction scores of the multiple choice classification head (scores for each choice before SoftMax). hidden_states (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_hidden_states=True`` is passed or when ``config.output_hidden_states=True``): Tuple of :obj:`tf.Tensor` (one for the output of the embeddings + one for the output of each layer) of shape :obj:`(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs. attentions (:obj:`tuple(tf.Tensor)`, `optional`, returned when ``output_attentions=True`` is passed or when ``config.output_attentions=True``): Tuple of :obj:`tf.Tensor` (one for each layer) of shape :obj:`(batch_size, num_heads, sequence_length, sequence_length)`. Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads. """ logits: tf.Tensor = None mc_logits: tf.Tensor = None hidden_states: Optional[Tuple[tf.Tensor]] = None attentions: Optional[Tuple[tf.Tensor]] = None OPENAI_GPT_START_DOCSTRING = r""" This model inherits from :class:`~transformers.TFPreTrainedModel`. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a `tf.keras.Model <https://www.tensorflow.org/api_docs/python/tf/keras/Model>`__ subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior. .. note:: TF 2.0 models accepts two formats as inputs: - having all inputs as keyword arguments (like PyTorch models), or - having all inputs as a list, tuple or dict in the first positional arguments. This second option is useful when using :meth:`tf.keras.Model.fit` method which currently requires having all the tensors in the first argument of the model call function: :obj:`model(inputs)`. If you choose this second option, there are three possibilities you can use to gather all the input Tensors in the first positional argument : - a single Tensor with :obj:`input_ids` only and nothing else: :obj:`model(inputs_ids)` - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: :obj:`model([input_ids, attention_mask])` or :obj:`model([input_ids, attention_mask, token_type_ids])` - a dictionary with one or several input Tensors associated to the input names given in the docstring: :obj:`model({"input_ids": input_ids, "token_type_ids": token_type_ids})` Parameters: config (:class:`~transformers.OpenAIGPTConfig`): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights. """ OPENAI_GPT_INPUTS_DOCSTRING = r""" Args: input_ids (:obj:`Numpy array` or :obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using :class:`~transformers.OpenAIGPTTokenizer`. See :func:`transformers.PreTrainedTokenizer.__call__` and :func:`transformers.PreTrainedTokenizer.encode` for details. `What are input IDs? <../glossary.html#input-ids>`__ attention_mask (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, sequence_length)`, `optional`): Mask to avoid performing attention on padding token indices. Mask values selected in ``[0, 1]``: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. `What are attention masks? <../glossary.html#attention-mask>`__ token_type_ids (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, sequence_length)`, `optional`): Segment token indices to indicate first and second portions of the inputs. Indices are selected in ``[0, 1]``: - 0 corresponds to a `sentence A` token, - 1 corresponds to a `sentence B` token. `What are token type IDs? <../glossary.html#token-type-ids>`__ position_ids (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, sequence_length)`, `optional`): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range ``[0, config.max_position_embeddings - 1]``. `What are position IDs? <../glossary.html#position-ids>`__ head_mask (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`): Mask to nullify selected heads of the self-attention modules. Mask values selected in ``[0, 1]``: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. inputs_embeds (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`): Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert :obj:`input_ids` indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (:obj:`bool`, `optional`): Whether or not to return the attentions tensors of all attention layers. See ``attentions`` under returned tensors for more detail. output_hidden_states (:obj:`bool`, `optional`): Whether or not to return the hidden states of all layers. See ``hidden_states`` under returned tensors for more detail. return_dict (:obj:`bool`, `optional`): Whether or not to return a :class:`~transformers.file_utils.ModelOutput` instead of a plain tuple. training (:obj:`bool`, `optional`, defaults to :obj:`False`): Whether or not to use the model in training mode (some modules like dropout modules have different behaviors between training and evaluation). """ @add_start_docstrings( "The bare OpenAI GPT transformer model outputting raw hidden-states without any specific head on top.", OPENAI_GPT_START_DOCSTRING, ) class TFOpenAIGPTModel(TFOpenAIGPTPreTrainedModel): def __init__(self, config, *inputs, **kwargs): super().__init__(config, *inputs, **kwargs) self.transformer = TFOpenAIGPTMainLayer(config, name="transformer") @add_start_docstrings_to_model_forward(OPENAI_GPT_INPUTS_DOCSTRING) @add_code_sample_docstrings( tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="openai-gpt", output_type=TFBaseModelOutput, config_class=_CONFIG_FOR_DOC, ) def call( self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, training=False, **kwargs, ): inputs = input_processing( func=self.call, config=self.config, input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, training=training, kwargs_call=kwargs, ) outputs = self.transformer( input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], token_type_ids=inputs["token_type_ids"], position_ids=inputs["position_ids"], head_mask=inputs["head_mask"], inputs_embeds=inputs["inputs_embeds"], output_attentions=inputs["output_attentions"], output_hidden_states=inputs["output_hidden_states"], return_dict=inputs["return_dict"], training=inputs["training"], ) return outputs # Copied from transformers.models.distilbert.modeling_tf_distilbert.TFDistilBertModel.serving_output def serving_output(self, output): hs = tf.convert_to_tensor(output.hidden_states) if self.config.output_hidden_states else None attns = tf.convert_to_tensor(output.attentions) if self.config.output_attentions else None return TFBaseModelOutput(last_hidden_state=output.last_hidden_state, hidden_states=hs, attentions=attns) @add_start_docstrings( """ OpenAI GPT Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings). """, OPENAI_GPT_START_DOCSTRING, ) class TFOpenAIGPTLMHeadModel(TFOpenAIGPTPreTrainedModel, TFCausalLanguageModelingLoss): def __init__(self, config, *inputs, **kwargs): super().__init__(config, *inputs, **kwargs) self.transformer = TFOpenAIGPTMainLayer(config, name="transformer") def get_output_embeddings(self): return self.get_input_embeddings() def set_output_embeddings(self, value): self.set_input_embeddings(value) @add_start_docstrings_to_model_forward(OPENAI_GPT_INPUTS_DOCSTRING) @add_code_sample_docstrings( tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="openai-gpt", output_type=TFCausalLMOutput, config_class=_CONFIG_FOR_DOC, ) def call( self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, labels=None, training=False, **kwargs, ): r""" labels (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`): Labels for computing the cross entropy classification loss. Indices should be in ``[0, ..., config.vocab_size - 1]``. """ inputs = input_processing( func=self.call, config=self.config, input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, labels=labels, training=training, kwargs_call=kwargs, ) transformer_outputs = self.transformer( input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], token_type_ids=inputs["token_type_ids"], position_ids=inputs["position_ids"], head_mask=inputs["head_mask"], inputs_embeds=inputs["inputs_embeds"], output_attentions=inputs["output_attentions"], output_hidden_states=inputs["output_hidden_states"], return_dict=inputs["return_dict"], training=inputs["training"], ) hidden_states = transformer_outputs[0] logits = self.transformer.tokens_embed(hidden_states, mode="linear") loss = None if inputs["labels"] is not None: # shift labels to the left and cut last logit token logits = logits[:, :-1] labels = inputs["labels"][:, 1:] loss = self.compute_loss(labels, logits) if not inputs["return_dict"]: output = (logits,) + transformer_outputs[1:] return ((loss,) + output) if loss is not None else output return TFCausalLMOutput( loss=loss, logits=logits, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) # Copied from transformers.models.bert.modeling_tf_bert.TFBertLMHeadModel.serving_output def serving_output(self, output): hs = tf.convert_to_tensor(output.hidden_states) if self.config.output_hidden_states else None attns = tf.convert_to_tensor(output.attentions) if self.config.output_attentions else None return TFCausalLMOutput(logits=output.logits, hidden_states=hs, attentions=attns) @add_start_docstrings( """ OpenAI GPT Model transformer with a language modeling and a multiple-choice classification head on top e.g. for RocStories/SWAG tasks. The two heads are two linear layers. The language modeling head has its weights tied to the input embeddings, the classification head takes as input the input of a specified classification token index in the input sequence). """, OPENAI_GPT_START_DOCSTRING, ) class TFOpenAIGPTDoubleHeadsModel(TFOpenAIGPTPreTrainedModel): def __init__(self, config, *inputs, **kwargs): super().__init__(config, *inputs, **kwargs) config.num_labels = 1 self.transformer = TFOpenAIGPTMainLayer(config, name="transformer") self.multiple_choice_head = TFSequenceSummary( config, initializer_range=config.initializer_range, name="multiple_choice_head" ) @add_start_docstrings_to_model_forward(OPENAI_GPT_INPUTS_DOCSTRING) @replace_return_docstrings(output_type=TFOpenAIGPTDoubleHeadsModelOutput, config_class=_CONFIG_FOR_DOC) def call( self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None, mc_token_ids=None, output_attentions=None, output_hidden_states=None, return_dict=None, training=False, **kwargs, ): r""" mc_token_ids (:obj:`tf.Tensor` or :obj:`Numpy array` of shape :obj:`(batch_size, num_choices)`, `optional`, default to index of the last token of the input): Index of the classification token in each input sequence. Selected in the range ``[0, input_ids.size(-1) - 1]``. Return: Examples:: >>> import tensorflow as tf >>> from transformers import OpenAIGPTTokenizer, TFOpenAIGPTDoubleHeadsModel >>> tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt') >>> model = TFOpenAIGPTDoubleHeadsModel.from_pretrained('openai-gpt') >>> # Add a [CLS] to the vocabulary (we should train it also!) >>> tokenizer.add_special_tokens({'cls_token': '[CLS]'}) >>> model.resize_token_embeddings(len(tokenizer)) # Update the model embeddings with the new vocabulary size >>> print(tokenizer.cls_token_id, len(tokenizer)) # The newly token the last token of the vocabulary >>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"] >>> encoding = tokenizer(choices, return_tensors="tf") >>> inputs = {k: tf.expand_dims(v, 0) for k, v in encoding.items()} >>> inputs["mc_token_ids"]= tf.constant([inputs["input_ids"].shape[-1] - 1, inputs["input_ids"].shape[-1] - 1])[None, :] # Batch size 1 >>> outputs = model(inputs) >>> lm_prediction_scores, mc_prediction_scores = outputs[:2] """ inputs = input_processing( func=self.call, config=self.config, input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, mc_token_ids=mc_token_ids, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, training=training, kwargs_call=kwargs, ) if inputs["input_ids"] is not None: input_shapes = shape_list(inputs["input_ids"]) else: input_shapes = shape_list(inputs["inputs_embeds"])[:-1] seq_length = input_shapes[-1] flat_input_ids = tf.reshape(inputs["input_ids"], (-1, seq_length)) if inputs["input_ids"] is not None else None flat_attention_mask = ( tf.reshape(inputs["attention_mask"], (-1, seq_length)) if inputs["attention_mask"] is not None else None ) flat_token_type_ids = ( tf.reshape(inputs["token_type_ids"], (-1, seq_length)) if inputs["token_type_ids"] is not None else None ) flat_position_ids = ( tf.reshape(inputs["position_ids"], (-1, seq_length)) if inputs["position_ids"] is not None else None ) transformer_outputs = self.transformer( flat_input_ids, flat_attention_mask, flat_token_type_ids, flat_position_ids, inputs["head_mask"], inputs["inputs_embeds"], inputs["output_attentions"], inputs["output_hidden_states"], return_dict=inputs["return_dict"], training=inputs["training"], ) hidden_states = transformer_outputs[0] hidden_states = tf.reshape(hidden_states, input_shapes + shape_list(hidden_states)[-1:]) lm_logits = self.transformer.tokens_embed(hidden_states, mode="linear") mc_logits = self.multiple_choice_head(hidden_states, inputs["mc_token_ids"], training=inputs["training"]) mc_logits = tf.squeeze(mc_logits, axis=-1) if not inputs["return_dict"]: return (lm_logits, mc_logits) + transformer_outputs[1:] return TFOpenAIGPTDoubleHeadsModelOutput( logits=lm_logits, mc_logits=mc_logits, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) @tf.function( input_signature=[ { "input_ids": tf.TensorSpec((None, None, None), tf.int32, name="input_ids"), "attention_mask": tf.TensorSpec((None, None, None), tf.int32, name="attention_mask"), "mc_token_ids": tf.TensorSpec((None, None), tf.int32, name="token_type_ids"), } ] ) def serving(self, inputs): output = self.call(inputs) return self.serving_output(output) def serving_output(self, output): hs = tf.convert_to_tensor(output.hidden_states) if self.config.output_hidden_states else None attns = tf.convert_to_tensor(output.attentions) if self.config.output_attentions else None return TFOpenAIGPTDoubleHeadsModelOutput( logits=output.logits, mc_logits=output.mc_logits, hidden_states=hs, attentions=attns ) @add_start_docstrings( """ The OpenAI GPT Model transformer with a sequence classification head on top (linear layer). :class:`~transformers.TFOpenAIGPTForSequenceClassification` uses the last token in order to do the classification, as other causal models (e.g. GPT-2) do. Since it does classification on the last token, it requires to know the position of the last token. If a :obj:`pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If no :obj:`pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the padding tokens when :obj:`inputs_embeds` are passed instead of :obj:`input_ids`, it does the same (take the last value in each row of the batch). """, OPENAI_GPT_START_DOCSTRING, ) class TFOpenAIGPTForSequenceClassification(TFOpenAIGPTPreTrainedModel, TFSequenceClassificationLoss): def __init__(self, config, *inputs, **kwargs): super().__init__(config, *inputs, **kwargs) self.num_labels = config.num_labels self.score = tf.keras.layers.Dense( config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="score", use_bias=False, ) self.transformer = TFOpenAIGPTMainLayer(config, name="transformer") @add_start_docstrings_to_model_forward(OPENAI_GPT_INPUTS_DOCSTRING) @add_code_sample_docstrings( tokenizer_class=_TOKENIZER_FOR_DOC, checkpoint="openai-gpt", output_type=TFSequenceClassifierOutput, config_class=_CONFIG_FOR_DOC, ) def call( self, input_ids=None, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None, inputs_embeds=None, output_attentions=None, output_hidden_states=None, return_dict=None, labels=None, training=False, **kwargs, ): r""" labels (:obj:`tf.Tensor` of shape :obj:`(batch_size, sequence_length)`, `optional`): Labels for computing the cross entropy classification loss. Indices should be in ``[0, ..., config.vocab_size - 1]``. """ inputs = input_processing( func=self.call, config=self.config, input_ids=input_ids, attention_mask=attention_mask, token_type_ids=token_type_ids, position_ids=position_ids, head_mask=head_mask, inputs_embeds=inputs_embeds, output_attentions=output_attentions, output_hidden_states=output_hidden_states, return_dict=return_dict, labels=labels, training=training, kwargs_call=kwargs, ) transformer_outputs = self.transformer( input_ids=inputs["input_ids"], attention_mask=inputs["attention_mask"], token_type_ids=inputs["token_type_ids"], position_ids=inputs["position_ids"], head_mask=inputs["head_mask"], inputs_embeds=inputs["inputs_embeds"], output_attentions=inputs["output_attentions"], output_hidden_states=inputs["output_hidden_states"], return_dict=inputs["return_dict"], training=inputs["training"], ) hidden_states = transformer_outputs[0] logits = self.score(hidden_states) logits_shape = shape_list(logits) in_logits = None if self.config.pad_token_id is None: sequence_lengths = -1 else: if inputs["input_ids"] is not None: sequence_lengths = ( tf.reduce_sum( tf.cast(tf.math.not_equal(inputs["input_ids"], self.config.pad_token_id), tf.int32), -1, keepdims=False, ) - 1 ) def get_seq_element(sequence_position, input_batch): return tf.strided_slice( input_batch, [sequence_position, 0], [sequence_position + 1, input_batch.shape[-1]], [1, 1] ) result = tf.map_fn( fn=lambda t: get_seq_element(t[0], t[1]), elems=[sequence_lengths, logits], dtype="float" ) in_logits = tf.reshape(result, [logits_shape[0], logits_shape[-1]]) else: sequence_lengths = -1 logger.warning( f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be " f"unexpected if using padding tokens in conjunction with `inputs_embeds.`" ) loss = None if inputs["labels"] is not None: if input_ids is not None: batch_size, sequence_length = shape_list(inputs["input_ids"])[:2] else: batch_size, sequence_length = shape_list(inputs["inputs_embeds"])[:2] assert ( self.config.pad_token_id is not None or batch_size == 1 ), "Cannot handle batch sizes > 1 if no padding token is defined." if not tf.is_tensor(sequence_lengths): in_logits = logits[0:batch_size, sequence_lengths] loss = self.compute_loss( tf.reshape(inputs["labels"], [-1, 1]), tf.reshape(in_logits, [-1, self.num_labels]) ) pooled_logits = in_logits if in_logits is not None else logits if not inputs["return_dict"]: output = (pooled_logits,) + transformer_outputs[1:] return ((loss,) + output) if loss is not None else output return TFSequenceClassifierOutput( loss=loss, logits=pooled_logits, hidden_states=transformer_outputs.hidden_states, attentions=transformer_outputs.attentions, ) # Copied from transformers.models.bert.modeling_tf_bert.TFBertForSequenceClassification.serving_output def serving_output(self, output): hs = tf.convert_to_tensor(output.hidden_states) if self.config.output_hidden_states else None attns = tf.convert_to_tensor(output.attentions) if self.config.output_attentions else None return TFSequenceClassifierOutput(logits=output.logits, hidden_states=hs, attentions=attns)

2024-01-10

Lucete28/TradeTrend

TT_runfile~update_naver_raw.py

from airflow.models.variable import Variable import openai import pandas as pd openai.api_key = Variable.get("gpt_api_key") Target_list = Variable.get("Target_list") values = [tuple(item.strip("()").split(",")) for item in Target_list.split("),")] values = [(x[0].strip(), x[1].strip()) for x in values] err_report = [] for val in values: gpt_ans = [] temp_df = pd.read_csv(f'/home/jhy/code/TradeTrend/data/{val[0]}/{val[0]}_temp4.csv') raw_df = pd.read_csv(f'/home/jhy/code/TradeTrend/data/{val[0]}/{val[0]}_news_raw2.csv') ans_list = raw_df.iloc[:, 1] while True: condition_satisfied = True # 모든 조건이 만족되었는지 여부를 추적하는 플래그 변수 for i, ans in enumerate(ans_list): try: if len(str(ans)) > 4 or (float(ans) > 1 or float(ans) < 0): messages = [] a = temp_df.iloc[i, 1] content = f'{a} {val[1]} 관련 뉴스기사 제목인데 {val[1]} 주식에 미칠 긍정도의 평균을 0에서 1사이 소숫점 두자리까지 나타내 float값만' messages.append({"role": "user", "content": content}) completion = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=messages ) chat_response = completion.choices[0].message.content gpt_ans.append(chat_response) messages.append({"role": "assistant", "content": chat_response}) # raw_df에서 해당 값을 새로운 값으로 업데이트합니다. raw_df.iloc[i, 1] = chat_response raw_df.to_csv(f'/home/jhy/code/TradeTrend/data/{val[0]}/{val[0]}_news_raw2.csv', index=False) condition_satisfied = False # 조건이 하나 이상의 항목에 대해 만족되지 않았음을 표시합니다. except: # 에러 발생 print(i, ans) err_report.append(ans) condition_satisfied = False if condition_satisfied: break # 모든 항목의 조건이 만족되었을 경우 반복문을 종료합니다. for err in err_report: if err_report.count(err) >=5: print("5회 이상 같은 err 발생") break

2024-01-10

LilithHafner/ai

integrated_ai.py

import openai openai.api_key = "sk-..." # GPT AI def ai(prompt): response = openai.Completion.create( engine="code-davinci-002", prompt=prompt, temperature=0, max_tokens=256, top_p=1, frequency_penalty=0, presence_penalty=0, stop="<end>" ) return response.choices[0].text # Subprocesses def user(prompt): return input(prompt+"\n*>> ") import traceback def python_eval(prompt): try: return str(eval(prompt, globals())) except: return traceback.format_exc() def python_exec(prompt): try: return str(exec(prompt, globals())) except: return traceback.format_exc() subprocesses = [ ("<user output>", "<user input>", user), ("<python eval>", "<python eval result>", python_eval), ("<python exec>", "<python exec result>", python_exec), ] def subprocess(s): for start, end, func in subprocesses: if s.startswith(start): return end + func(s[len(start):]) # print("The AI made an unsupported query:", s, "", sep="\n") return "<error>unknown tag" ## Training data prompt = """This is a question and answer bot that has oracles to various external tools including python, google, and others <user input>what time is it<end> <pyhton eval>time.ctime()<end> <python eval result>Traceback (most recent call last): File "/Users/x/Documents/integrated_ai.py", line 26, in python return str(eval(prompt, globals(), locals())) File "<string>", line 1, in <module> NameError: name 'time' is not defined<end> <python exec>import time<end> <python exec result>None<end> <python eval>time.ctime()<end> <user output>The time is Sun Apr 24 18:01:32 2022<end> <user input>what is the weather in New York<end> <google>weather in New York<end> <google result>Sunny 53°F°C Precipitation: 1% Humidity: 52% Wind: 7 mph New York, NY Sunday 6:00 PM Sunny TemperaturePrecipitationWind<end> <user output>The weather in New York is Sunny<end> <user input>is it warm in chicago?<end> <google>weather in chicago<end> result: Cloudy 70°F°C Precipitation: 5% Humidity: 65% Wind: 19 mph Chicago, IL Sunday 6:00 PM Cloudy TemperaturePrecipitationWind<end> <user output>It is warm in chicago<end> <user input>is 1729 prime?<end> <python eval>is_prime(1729)<end> <python eval result>Traceback (most recent call last): File "/Users/x/Documents/integrated_ai.py", line 26, in python_eval return str(eval(prompt, globals())) File "<string>", line 1, in <module> NameError: name 'is_prime' is not defined<end> <python exec>def is_prime(n): if n <= 1: return False for i in range(2, n): if n % i == 0: return False return True<end> <python exec result>None<end> <python eval>is_prime(1729)<end> <python eval result>False<end> <user output>1729 is not prime<end> <user input>Stop using google<end> <user output>Google disabled.<end> <user input>What's the weather?<end> <user output>I cannot answer that question without google<end> <user input>Name 7 edibe mushrooms<end> <user output>Pleurotus, Lentinula edodes, Shiitake mushroom, Auricularia auricula-judae, Volvariella volvacea, Flammulina velutipes, Tremella fuciformis<end>""" # Main loop def kernal(verbose=True): global prompt prompt += "<user input>" + user("Welcome!") + "<end>\n" while True: call = ai(prompt) if verbose: print(call + "<end>") prompt += call + "<end>\n" if call.startswith("<exit>"): return result = subprocess(call) if verbose: print(result + "<end>") prompt += result + "<end>\n" if __name__ == "__main__": kernal()

2024-01-10

Kororinpas/Lit_Tool

document_util.py

def get_split_documents(docs, chunk_size, chunk_overlap): from langchain.text_splitter import RecursiveCharacterTextSplitter text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size,chunk_overlap=chunk_overlap) return text_splitter.split_documents(docs)

2024-01-10

Kororinpas/Lit_Tool

literature_test.py

import streamlit as st import sys class StreamlitWriter: def write(self, text): st.write(text.strip()) ### This the function about streamlit def Vector_Databse(): st.write("Vector Database") choose = st.radio("Choose using an existing database or upload a new one.", ["Using an existing one", "Uploading a new one"]) import torch device = 'cuda' if torch.cuda.is_available() else 'cpu' if choose == "Using an existing one": persist_dirctory = st.text_input("Enter the persist_dirctory") collection = st.text_input("Enter the collection") if st.button('Confirm'): st.session_state['persist_dirctory'] = persist_dirctory st.session_state['collection'] = collection vectorstore,embeddings = load_vectorstore(persist_directory=st.session_state['persist_dirctory'], collection_name = st.session_state['collection'], model_name = 'sentence-transformers/all-mpnet-base-v2', device = device) st.session_state['vectorstore'] = vectorstore st.session_state['embeddings'] = embeddings print('The vectorstore load successfully') else: path = st.text_input("Enter the path") persist_dirctory = st.text_input("Enter the persist_dirctory") collection = st.text_input("Enter the collection") if st.button('Confirm'): st.session_state['path'] = path st.session_state['persist_dirctory'] = persist_dirctory st.session_state['collection'] = collection split_docs = load_pdf(path = st.session_state['path'], openai_api_key=st.session_state['openai_api_key'], chunk_size=st.session_state['chunk_size'], chunk_overlap=st.session_state['chunk_overlap']) vectorstore,embeddings = generate_vectorstore(split_docs = split_docs, model_name = 'sentence-transformers/all-mpnet-base-v2', persist_directory = st.session_state['persist_dirctory'], collection_name = st.session_state['collection'], device=device) st.session_state['vectorstore'] = vectorstore st.session_state['embeddings'] =embeddings print('The vectorstore load successfully') def Parameters(): import os openai_api_key = st.text_input('Enter your Openapi_api_key') if st.button('Confirm'): if openai_api_key == '': st.session_state['openai_api_key'] = os.environ.get('openai_api_key') else: st.session_state['openai_api_key'] = openai_api_key chunk_size = st.text_input('Enter your chunk_size') if st.button('Confirm_1'): if chunk_size== '': st.session_state['chunk_size'] = 1500 chunk_overlap = st.text_input('Enter your chunk_overlap') if st.button('Confirm_2'): if chunk_overlap == '': st.session_state['chunk_overlap'] = 0 def Docs(): col1,col2 = st.columns([1,1]) with col1: output_text = '' vectorstore = st.session_state['vectorstore'] edited_output_text = st.text_area("输出文本", value=output_text, height=600) if st.button("Confirm paragraph"): output_text = edited_output_text k = st.slider("Select the number of sentences to generate", min_value=1, max_value=5, value=1) query = st.text_input("Input the query") if st.button("Confirm query"): output, docs = get_chain_output(query=query, vectordb=vectorstore, k=k, openai_api_key=st.session_state['openai_api_key']) final_json = run_text_match(output=output, query=query, docs=docs, k=k, embeddings=st.session_state['embeddings']) st.session_state['final_json'] = final_json with col2: if 'final_json' in st.session_state: final_json = st.session_state['final_json'] selected_sentence = st.selectbox("Select a sentence", final_json) if st.button('Confirm sentence'): process_selected_sentence(selected_sentence) ###This is the function about Langchain ###Loading PDF part def load_pdf(path, openai_api_key, chunk_size, chunk_overlap): from langchain.document_loaders import PyMuPDFLoader, DirectoryLoader, UnstructuredPDFLoader #from detectron2.config import get_cfg from PyPDF2 import PdfReader #cfg = get_cfg() #cfg.MODEL.DEVICE = 'gpu' import os file_names = os.listdir(path) pdf_file_names = [path + '/' + file_name for file_name in file_names if file_name.endswith('.pdf')] docs = [] import re for pdf in pdf_file_names: source = extract_doi(pdf) if source != 'None': doc = PyMuPDFLoader(pdf).load() for element in doc: element.metadata = source element.page_content = re.sub('\n+', ' ', element.page_content.strip()) docs.append(element) else: doc = PyMuPDFLoader(pdf).load() print(f"{pdf} is not identified! Using other strategy!!") source = extract_doi_llm(doc, openai_api_key) if source != 'None': for element in doc: element.metadata = source for element in doc: element.page_content = re.sub('\n+', ' ', element.page_content.strip()) docs.append(element) from langchain.text_splitter import RecursiveCharacterTextSplitter text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap) split_docs = text_splitter.split_documents(docs) return split_docs def get_info(path): from PyPDF2 import PdfReader with open(path, 'rb') as f: pdf = PdfReader(f) info = pdf.metadata return info def extract_doi(path): source = 0 info = get_info(path) if '/doi' in info: doi = info['/doi'] elif '/Subject' in info: Subject = info['/Subject'] if 'doi:' in Subject: Subject = Subject.split('doi:') doi = Subject[1] else: source = 'None' elif '/WPS-ARTICLEDOI' in info: doi = info['/WPS-ARTICLEDOI'] else: source = 'None' if source != 'None': import habanero import time citation = habanero.cn.content_negotiation(ids=doi, format='bibentry') time.sleep(5) import bibtexparser citation = bibtexparser.loads(citation) citation = citation.entries[0] source = {'author': citation['author'], 'year': citation['year'], 'title': citation['title'], 'journal': citation['journal'], } return source def extract_doi_llm(doc,openai_api_key): import re doc[0].page_content = re.sub('\n+',' ',doc[0].page_content.strip()) from langchain.text_splitter import RecursiveCharacterTextSplitter text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500,chunk_overlap = 50) split_docs = text_splitter.split_documents(doc) abstract = split_docs[0] doi = extract_chain(abstract,openai_api_key) if doi != 'None' and doi!= None: import habanero import time citation = habanero.cn.content_negotiation(ids = doi,format='bibentry') time.sleep(5) import bibtexparser citation = bibtexparser.loads(citation) citation = citation.entries[0] source = {'author':citation['author'], 'year':citation['year'], 'title':citation['title'], 'journal':citation['journal'], } return source else: source = 'None' return source def extract_chain(abstract, openai_api_key): from kor.extraction import create_extraction_chain from kor.nodes import Object, Text, Number from langchain.chat_models import ChatOpenAI llm = ChatOpenAI( model_name="gpt-3.5-turbo", openai_api_key=openai_api_key, temperature=0, ) schema = Object( id="doi", description="doi is a digital identifier.It typically starts with 10. followed by a numeric prefix, such as 10.1000/182.", attributes=[ Text( id="doi", description='doi is a digital identifier. It typically starts with "10." followed by a numeric prefix, such as 10.1000/182.', examples=[ ( 'American Economic Journal: Economic Policy 2015, 7(4): 223–242 http://dx.doi.org/10.1257/pol.20130367 223 Water Pollution Progress at Borders: The', 'http://dx.doi.org/10.1257/pol.20130367'), ( 'Environment and Development Economics (2020), 1–17 doi:10.1017/S1355770X2000025X EDE RESEARCH ARTICLE Political incentives, Party Congress, and pollution cycle: empirical evidence from China Zhihua Tian,1 and Yanfang Tian2* 1School of Economics, Zhejiang University of Technology, Hangzhou', '10.1017/S1355770X2000025X') ], many=True ) ], many=False ) chain = create_extraction_chain(llm, schema, encoder_or_encoder_class='json') output = chain.predict_and_parse(text=abstract.page_content) if 'doi' not in output['data']: print(f"LLM strategy failed!!{abstract.metadata['source']} Please manually add it!!") source = 'None' return source else: if output['data']['doi']['doi'] == []: print(f"LLM strategy failed!!{abstract.metadata['source']} Please manually add it!!") source = 'None' return source else: doi = output['data']['doi']['doi'][0] if 'doi=' in doi: doi = doi.split('doi=')[1] return doi ###Loading the database def generate_vectorstore(split_docs, device, model_name, persist_directory, collection_name): from langchain.vectorstores import Chroma from langchain.embeddings import HuggingFaceEmbeddings model_kwargs = {'device': device} model_name = model_name embeddings = HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs) persist_directory = persist_directory collection_name = collection_name vectorstore = Chroma.from_documents(split_docs, embeddings, collection_name=collection_name, persist_directory=persist_directory) vectorstore.persist() return vectorstore,embeddings def load_vectorstore(persist_directory,device,model_name,collection_name): from langchain.vectorstores import Chroma from langchain.embeddings import HuggingFaceEmbeddings model_kwargs = {'device': device} model_name = model_name embeddings = HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs) vectordb = Chroma(collection_name=collection_name, persist_directory=persist_directory, embedding_function=embeddings) return vectordb,embeddings ###Using Langchain and match def get_chain_output(query, vectordb, k, openai_api_key): docs = vectordb.similarity_search(query, 6, include_metadata=True) from langchain.chat_models import ChatOpenAI llm = ChatOpenAI(openai_api_key=openai_api_key, temperature=0, model_name="gpt-3.5-turbo") from langchain.prompts import PromptTemplate, ChatPromptTemplate, HumanMessagePromptTemplate from langchain.llms import OpenAI from langchain.output_parsers import PydanticOutputParser from pydantic import BaseModel, Field, validator from typing import List, Union, Optional class Sentence(BaseModel): sentence: List[str] = Field( description="The sentence in the given document which is the most similar to the query provided") source: List[str] = Field(description="The meta source of the paper") score: List[float] = Field( description="The similarity score between the sentence selected and the query provided") parser = PydanticOutputParser(pydantic_object=Sentence) dic = {'1':"one", "2":"two", "3":"three", "4":"four", "5":"five"} k = dic[str(k)] question_template = f""" Given the document and query, find {k} sentences in the document that are most similar in meaning to the query. Return the sentences, the meta source of the sentences and the cosine similarity scores. If no similar sentences is found, return the sentence with highest cosine siliarity scores. """ main_template = """ {query} =========== {context} =========== {format_instructions} """ question_template = question_template+main_template from langchain.chains.question_answering import load_qa_chain from langchain import LLMChain PROMPT = PromptTemplate(template=question_template, input_variables=['query', 'context'], partial_variables={"format_instructions": parser.get_format_instructions()}) llm_chain = LLMChain(llm=llm, prompt=PROMPT) output = llm_chain({"query": query, "context": docs}) return output, docs def run_text_match(output, k,query, docs,embeddings): import re text = re.sub("\n+", "", output['text']) import json json_obj = json.loads(text) if "properties" in json_obj: print('No result was found, Using embedding searching strategy!!!') split_docs = split_for_embedding(docs) similar_sentence = search_cosine_similarity(query,k,split_docs, embeddings) return similar_sentence else: json_obj = [{'sentence': json_obj['sentence'][i], 'source': json_obj['source'][i], 'score': json_obj['score'][i]} for i in range(k)] return json_obj def split_for_embedding(docs): ##输入docs(list),输出split_for embedding(list) for_embedding = [] for content in docs: new_content = content.page_content.replace('et al.', 'et al。') new_content = new_content.split('.') if 'source' in content.metadata: meta_data = content.metadata['source'] else: meta_data = content.metadata for split_content in new_content: split_content = split_content.replace('。', '.') if len(split_content) < 30: continue else: for_embedding.append({"content": split_content, "source": meta_data}) return for_embedding def search_cosine_similarity(query, k,split_docs, embeddings): ##query-str,split_docs-list,embeddings-embeddings() split_docs_content = [content['content'] for content in split_docs] split_docs_content = list(set(split_docs_content)) embed_docs = embeddings.embed_documents(split_docs_content) embed_query = embeddings.embed_query(query) from openai.embeddings_utils import cosine_similarity cos_index = [] for embed_doc in embed_docs: cos_index.append(cosine_similarity(embed_doc, embed_query)) # 这边是根据大小建立索引 idx = sorted(range(len(cos_index)), key=lambda k: cos_index[k]) # 根据cos_index的大小进行排序 final_similar_list = [] for index in idx[-k:]: unit = {} unit['sentences'] = split_docs_content[index] unit['source'] = split_docs[index]['source'] unit['score'] = cos_index[index] final_similar_list.append(unit) return final_similar_list def main(): st.title("Literature Review Tool") sys.stdout = StreamlitWriter() # Create a toggle button to switch between pages page = st.sidebar.radio("Choose a page", [ "Parameter","Vector Database","Docs"]) if page == "Parameter": Parameters() elif page == "Vector Database": Vector_Databse() elif page == "Docs": Docs() def my_function(input_text): # 在此处添加您的处理逻辑 output_text = input_text.upper() return output_text def process_selected_sentence(selected_sentence): # 在最终输出区域展示用户选择的句子 st.write(f"You selected: {selected_sentence}") main()

2024-01-10

Kororinpas/Lit_Tool

pdf_retrieval.py

from operator import itemgetter from langchain.chat_models import ChatOpenAI from langchain.output_parsers import StructuredOutputParser, ResponseSchema from langchain.prompts import ChatPromptTemplate, HumanMessagePromptTemplate from langchain.document_loaders import DataFrameLoader, PyMuPDFLoader import os import fitz import pandas as pd import json import ast def fonts(doc, granularity=False, pages=2): """Extracts fonts and their usage in PDF documents. :param doc: PDF document to iterate through :type doc: <class 'fitz.fitz.Document'> :param granularity: also use 'font', 'flags' and 'color' to discriminate text :type granularity: bool :rtype: [(font_size, count), (font_size, count}], dict :return: most used fonts sorted by count, font style information """ styles = {} font_counts = {} pageCounter = 0 for page in doc: blocks = page.get_text("dict")["blocks"] for b in blocks: # iterate through the text blocks if b['type'] == 0: # block contains text for l in b["lines"]: # iterate through the text lines for s in l["spans"]: # iterate through the text spans if granularity: identifier = "{0}_{1}_{2}_{3}".format(s['size'], s['flags'], s['font'], s['color']) styles[identifier] = {'size': s['size'], 'flags': s['flags'], 'font': s['font'], 'color': s['color']} else: identifier = "{0}".format(s['size']) styles[identifier] = {'size': s['size'], 'font': s['font']} font_counts[identifier] = font_counts.get(identifier, 0) + 1 # count the fonts usage pageCounter += 1 if pageCounter >= pages: break font_counts = sorted(font_counts.items(), key=itemgetter(1), reverse=True) if len(font_counts) < 1: raise ValueError("Zero discriminating fonts found!") return font_counts, styles def font_tags(font_counts, styles): """Returns dictionary with font sizes as keys and tags as value. :param font_counts: (font_size, count) for all fonts occuring in document :type font_counts: list :param styles: all styles found in the document :type styles: dict :rtype: dict :return: all element tags based on font-sizes """ p_style = styles[font_counts[0][0]] # get style for most used font by count (paragraph) p_size = p_style['size'] # get the paragraph's size # sorting the font sizes high to low, so that we can append the right integer to each tag font_sizes = [] for (font_size, count) in font_counts: font_sizes.append(float(font_size)) font_sizes.sort(reverse=True) # aggregating the tags for each font size idx = 0 size_tag = {} for size in font_sizes: idx += 1 if size == p_size: idx = 0 size_tag[size] = '<p>' if size > p_size: size_tag[size] = '<h{0}>'.format(idx) elif size < p_size: size_tag[size] = '<s{0}>'.format(idx) return size_tag def get_pdf_raw_pages(doc, pages): header_para = [] pageCounter = 0 for page in doc: blocks = page.get_text("dict")["blocks"] header_para.append(blocks) pageCounter += 1 if pageCounter >= pages: break return header_para def headers_para(doc, size_tag, pages=2): """Scrapes headers & paragraphs from PDF and return texts with element tags. :param doc: PDF document to iterate through :type doc: <class 'fitz.fitz.Document'> :param size_tag: textual element tags for each size :type size_tag: dict :rtype: list :return: texts with pre-prended element tags """ header_para = [] # list with headers and paragraphs first = True # boolean operator for first header previous_s = {} # previous span pageCounter = 0 for page in doc: blocks = page.get_text("dict")["blocks"] for b in blocks: # iterate through the text blocks # header_para.append("<section_block>") if b['type'] == 0: # this block contains text # REMEMBER: multiple fonts and sizes are possible IN one block block_string = "" # text found in block for l in b["lines"]: # iterate through the text lines for s in l["spans"]: # iterate through the text spans if s['text'].strip(): # removing whitespaces: if first: previous_s = s first = False block_string = size_tag[s['size']] + s['text'] else: if s['size'] == previous_s['size']: if block_string and all((c == "|") for c in block_string): # block_string only contains pipes block_string = size_tag[s['size']] + s['text'] if block_string == "": # new block has started, so append size tag block_string = size_tag[s['size']] + s['text'] else: # in the same block, so concatenate strings block_string += " " + s['text'] else: header_para.append(block_string) block_string = size_tag[s['size']] + s['text'] previous_s = s # new block started, indicating with a pipe block_string += "|" # header_para.append("<text_block>") header_para.append(block_string) # header_para.append("<text_block_end>") # header_para.append("<section_block_end>") pageCounter += 1 if pageCounter >= pages: break return header_para def get_pdf_first_page_txt(pdf_path, pages=2): doc = fitz.open(pdf_path) font_counts, styles = fonts(doc, granularity=False, pages=pages) size_tag = font_tags(font_counts, styles) return headers_para(doc, size_tag, pages) def get_pdf_pages(pdf_path, pages=2): docs = PyMuPDFLoader(pdf_path).load() return docs[:pages] # texts = [] # for doc in docs[:pages]: # texts.append(doc.page_content) # return texts def get_pdf_page_metadata(pdf_path, pages): pdf_first_page_txt = get_pdf_first_page_txt(pdf_path, pages) template = """ I have extracted text from the initial pages of a Journal of Economic Literature (JEL) PDF file. I require assistance in extracting specific details, namely: article title, author, abstract and keywords section. Please be aware that if you encounter JEL classifications such as C12 and P34, kindly ignore them and refrain from including them in the abstract and keywords. {format_instructions} Wrap your final output as a json objects INPUT: {pdf_first_page_txt} YOUR RESPONSE: """ response_schemas = [ ResponseSchema(name="title", description="extracted title"), ResponseSchema(name="author", description="extracted authors seperated by comma"), ResponseSchema(name="abstract", description="extracted abstract"), ResponseSchema(name="keywords", description="extracted keywords") ] output_parser = StructuredOutputParser.from_response_schemas(response_schemas) prompt = ChatPromptTemplate( messages=[ HumanMessagePromptTemplate.from_template(template) ], input_variables=["pdf_first_page_txt"], partial_variables={"format_instructions": output_parser.get_format_instructions()} ) llm = ChatOpenAI(model_name='gpt-3.5-turbo-16k',temperature=0.0,max_tokens=6048) # type: ignore gpt-3.5-turbo final_prompt = prompt.format_prompt(pdf_first_page_txt=pdf_first_page_txt) output = llm(final_prompt.to_messages()) try: result = output_parser.parse(output.content) except: if "```json" in output.content: json_string = output.content.split("```json")[1].strip() else: json_string = output.content result = fix_JSON(json_string) head, tail = os.path.split(pdf_path) result["filename"] = tail return result def get_pdf_page_accept_metadata(pdf_path, pages): pdf_first_page_txt = get_pdf_first_page_txt(pdf_path, pages) template = """ I have extracted text from the initial pages of a Journal of Economic Literature (JEL) PDF file. I need help identifying the accepted date of the article. If the accepted date is not explicitly specified, it should be located either at the top or bottom of the first or second page of the article in a date format without the prefix 'accepted'. {format_instructions} Wrap your final output as a json objects INPUT: {pdf_first_page_txt} YOUR RESPONSE: """ response_schemas = [ ResponseSchema(name="accepted", description="extracted accepted date") ] output_parser = StructuredOutputParser.from_response_schemas(response_schemas) prompt = ChatPromptTemplate( messages=[ HumanMessagePromptTemplate.from_template(template) ], input_variables=["pdf_first_page_txt"], partial_variables={"format_instructions": output_parser.get_format_instructions()} ) llm = ChatOpenAI(model_name='gpt-3.5-turbo',temperature=0.0,max_tokens=148) # type: ignore gpt-3.5-turbo final_prompt = prompt.format_prompt(pdf_first_page_txt=pdf_first_page_txt) output = llm(final_prompt.to_messages()) try: result = output_parser.parse(output.content) except: if "```json" in output.content: json_string = output.content.split("```json")[1].strip() else: json_string = output.content result = fix_JSON(json_string) head, tail = os.path.split(pdf_path) result["filename"] = tail return result def get_pdf_intro(pdf_path, pages): pdf_first_page_txt = get_pdf_first_page_txt(pdf_path, pages) template = """ I have extracted text from the initial pages of a Journal of Economic Literature (JEL) PDF file. I require assistance in extracting introduction section. Typically, the introduction section begins after the abstract and ends before the next sub-title or section heading. Wrap your final output as a json objects INPUT: {pdf_first_page_txt} YOUR RESPONSE: """ response_schemas = [ ResponseSchema(name="introduction", description="extracted introduction") ] output_parser = StructuredOutputParser.from_response_schemas(response_schemas) prompt = ChatPromptTemplate( messages=[ HumanMessagePromptTemplate.from_template(template) ], input_variables=["pdf_first_page_txt"], # partial_variables={"format_instructions": output_parser.get_format_instructions()} ) llm = ChatOpenAI(model_name='gpt-3.5-turbo-16k',temperature=0.0,max_tokens=8396) # type: ignore gpt-3.5-turbo final_prompt = prompt.format_prompt(pdf_first_page_txt=pdf_first_page_txt) output = llm(final_prompt.to_messages()) try: result = output_parser.parse(output.content) except Exception as e: print(str(e)) if "```json" in output.content: json_string = output.content.split("```json")[1].strip() else: json_string = output.content result = fix_JSON(json_string) head, tail = os.path.split(pdf_path) result["filename"] = tail return result def get_polish_intro(my_intro, sample_introes, words_limit, temperature): template = """ I require an introduction for my Journal of Economic Literature and I would appreciate it \ if you could compose it for me around {words_limit} words. I would like the introduction mimic on the \ sample introductions that I will provide. If I have already provided my own introduction, \ please refine it accordingly. % My own introduction: {my_intro} % Sample introductions: {sample_introes} % End of sample introductions: YOUR RESPONSE: """ response_schemas = [ ResponseSchema(name="introduction", description="refined introduction") ] output_parser = StructuredOutputParser.from_response_schemas(response_schemas) prompt = ChatPromptTemplate( messages=[ HumanMessagePromptTemplate.from_template(template) ], input_variables=["my_intro","sample_introes","words_limit"], partial_variables={"format_instructions": output_parser.get_format_instructions()} ) llm = ChatOpenAI(model_name='gpt-3.5-turbo',temperature=temperature,max_tokens=2048) # type: ignore gpt-3.5-turbo final_prompt = prompt.format_prompt(my_intro=my_intro, sample_introes=sample_introes, words_limit=words_limit) output = llm(final_prompt.to_messages()) result = output.content return result def fix_JSON(json_message=None): result = None try: result = json.loads(json_message) except Exception as e: # Find the offending character index: idx_to_replace = int(str(e).split(' ')[-1].replace(')', '')) # Remove the offending character: json_message = list(json_message) json_message[idx_to_replace] = ' ' new_message = ''.join(json_message) return fix_JSON(json_message=new_message) return result def save_pdfs_to_db(pdf_files, excel_file, meta_type='meta', pages=2): if os.path.exists(excel_file): df = pd.read_excel(excel_file) existing_data = df.to_dict(orient='records') else: existing_data = [] existing_filenames = set(row['filename'] for row in existing_data) for doc in pdf_files: head, tail = os.path.split(doc) if tail not in existing_filenames: # print('get meta from LLM '+doc) try: if meta_type == 'intro': metadata = get_pdf_intro2(doc, pages) elif meta_type == 'date': metadata = get_pdf_page_accept_metadata(doc, pages) else: metadata = get_pdf_page_metadata(doc, pages) temp_data = [] temp_data.append(metadata) save_to_excel(existing_data+temp_data, excel_file) existing_data += temp_data print("Data append to ", excel_file) except Exception as e: print(str(e)) def get_metadata_from_db(excel_file): df = pd.read_excel(excel_file) dict = df.to_dict(orient='records',) return dict def get_column_from_db(excel_file, column): df = pd.read_excel(excel_file) doc = DataFrameLoader(df, column).load() return doc def get_data_from_csv(file_path, column_name, filter_value): data = pd.read_csv(file_path, encoding = 'unicode_escape') filtered_data = data[data[column_name] == filter_value] dict_data = filtered_data.to_dict(orient='records') #filtered_data.values.tolist() for row in dict_data: md = ast.literal_eval(row["metadata"]) # print(type(md)) row["date"] = md["modDate"] return dict_data def get_filename_list(similar_dict, path): filenames = [] for doc in similar_dict['context']: filenames.append(os.path.join(path, doc.metadata['filename'])) return filenames def save_to_excel(data, file_path): df = pd.DataFrame(data) df.to_excel(file_path, index=False) def get_pdf_intro2(pdf_path, pages): pdf_first_page_txt = get_pdf_first_page_txt(pdf_path, pages) # pdf_first_page_txt = get_pdf_pages(pdf_path, pages) human_template = """ I have extracted the text from the initial pages of a Journal of Economic Literature (JEL) PDF file. I require assistance in extracting the introduction section. Typically, the document follows a pattern where the 'abstract' header is encountered, followed by the abstract section. Subsequently, an 'Introduction' header is expected, which is followed by the introduction section. Next, there may be a 'Background' header or other headers indicating different sections. The introduction section generally concludes before the next sub-title or section heading appears, such as 'Background' or other similar headings. Please continue searching for the introduction section until you reach a clear next sub-title or section heading. However, please note that if you encounter a bottom part between two pages, such as a section starting with 'RECEIVED:' followed by a date, it does not necessarily mean that the introduction section has ended. In such cases, you should continue searching on the next page. If the text 'www.elsevier.com' appears in the beginning, it indicates that the literature is published on Elsevier and follows a specific format. In this case, the abstract section will start with "A B S T R A C T" and end before the introduction section. The introduction section will typically start with "1. Introduction" and end before the next section header, such as "2. Background". Please continue searching for the introduction section until you reach next section heading such as "2. Background", it has to be started with "2.". Please provide the introduction section as the final output in JSON format with the key 'Introduction' written in Pascal case. Exclude the content of the abstract section. Only include the text within the introduction section and exclude any text prior to it. INPUT: {pdf_first_page_txt} YOUR RESPONSE: """ response_schemas = [ # ResponseSchema(name="abstract", description="extracted abstract"), ResponseSchema(name="introduction", description="extracted introduction") ] output_parser = StructuredOutputParser.from_response_schemas(response_schemas) prompt = ChatPromptTemplate( messages=[ HumanMessagePromptTemplate.from_template(human_template) ], input_variables=["pdf_first_page_txt"] ) llm = ChatOpenAI(model_name='gpt-3.5-turbo-16k',temperature=0.0,max_tokens=6658) # type: ignore gpt-3.5-turbo final_prompt = prompt.format_prompt(pdf_first_page_txt=pdf_first_page_txt) output = llm(final_prompt.to_messages()) try: result = output_parser.parse(output.content) except Exception as e: print(str(e)) if "```json" in output.content: json_string = output.content.split("```json")[1].strip() else: json_string = output.content result = fix_JSON(json_string) head, tail = os.path.split(pdf_path) result["filename"] = tail return result def main(): documents = ['./data/docs/literature/Do people care about democracy_An experiment exploring the value of voting rights.pdf', './data/docs/literature/Expressive voting versus information avoidance_expenrimental evidence in the context of climate change mitigation.pdf', './data/docs/literature/Crashing the party_An experimental investigation of strategic voting in primary elections.pdf', './data/docs/literature/Economic growth and political extremism.pdf'] doc = './data/docs/literature_suicide/1-s2.0-S0304387821000432-main.pdf' doc = './data/docs/literature_suicide/1-s2.0-S0047272721000761-main.pdf' # doc = './data/docs/literature_suicide/rest_a_00777.pdf' documents = ['./data/docs/literature/Do people care about democracy_An experiment exploring the value of voting rights.pdf' ,'./data/docs/literature/Expressive voting versus information avoidance_expenrimental evidence in the context of climate change mitigation.pdf' ,'./data/docs/literature/Economic growth and political extremism.pdf' ] # './data/docs/literature/Expressive voting versus information avoidance_expenrimental evidence in the context of climate change mitigation.pdf', # './data/docs/literature/Crashing the party_An experimental investigation of strategic voting in primary elections.pdf', # './data/docs/literature/Economic growth and political extremism.pdf'] # save_pdfs_to_db(documents, intro_excel_file, is_Intro=True, pages=4) metadata = get_pdf_intro2(doc, 2) print(metadata) # docs = get_pdf_first_page_txt(doc, 3) # # docs = get_pdf_pages(doc, 2) # # docs = get_pdf_raw_pages(doc, 2) # print(docs) # pdf_first_page_txt = get_pdf_first_page_txt(doc, 3) # raw_txt = get_pdf_raw_pages(fitz.open(doc), 2) # print(raw_txt) # pdf_first_page_txt = get_pdf_first_page_txt(doc, 3) # output_file = "data/db/repo_intro_4.xlsx" # intro354_excel_file = "data/db/repo_intro_35_16.xlsx" # save_pdfs_to_db(documents, intro354_excel_file, is_intro=True, pages=4) # intros = [dict["introduction"] for dict in get_metadata_from_db(intro35_excel_file)] # polish = get_polish_intro('', intros[:3], 600, 0) # print(polish) # csv_file = "./data/db/summary.csv" # column_name = "Theme" # filter_value = "China authoritarian system" # data = get_data_from_csv(csv_file, column_name, filter_value) # print(data) if __name__ == '__main__': main()

2024-01-10

Kororinpas/Lit_Tool

pdf_documents.py

from pdf_metadata import get_pdf_metadata from pdf_metadata_llm import get_pdf_metadata_using_llm def get_pdf_documents(pdf_files): from langchain.document_loaders import PyMuPDFLoader,DirectoryLoader,UnstructuredPDFLoader docs =[] import re for pdf_fullpath in pdf_files: metadata = get_pdf_metadata(pdf_fullpath) if metadata != 'None': doc = PyMuPDFLoader(pdf_fullpath).load() for element in doc: element.metadata = metadata element.page_content = re.sub('\n+',' ',element.page_content.strip()) docs.append(element) else: doc = PyMuPDFLoader(pdf_fullpath).load() print(f"{pdf_fullpath} is not identified! Using other strategy!!") metadata = get_pdf_metadata_using_llm(doc) if metadata != 'None': for element in doc: element.metadata = metadata for element in doc: element.page_content = re.sub('\n+',' ',element.page_content.strip()) docs.append(element) return docs

2024-01-10

Kororinpas/Lit_Tool

pdf_metadata_llm.py

from doi import get_doi from document_util import get_split_documents def get_pdf_metadata_using_llm(doc): import re doc[0].page_content = re.sub('\n+',' ',doc[0].page_content.strip()) # from langchain.text_splitter import RecursiveCharacterTextSplitter # text_splitter = RecursiveCharacterTextSplitter(chunk_size=1500,chunk_overlap = 50) split_docs = get_split_documents(doc, 1500, 50) abstract = split_docs[0] doi = get_doi(abstract) if doi != 'None': import habanero import time citation = habanero.cn.content_negotiation(ids = doi,format='bibentry') time.sleep(5) import bibtexparser citation = bibtexparser.loads(citation) citation = citation.entries[0] metadata = {'author':citation['author'], 'year':citation['year'], 'title':citation['title'], 'journal':citation['journal'], } return metadata else: metadata = 'None' return metadata

2024-01-10

Kororinpas/Lit_Tool

cosine_match.py

def search_cosine_similarity(query,split_docs,embeddings): ##query-str,split_docs-list,embeddings-embeddings() split_docs_content = [content['content'] for content in split_docs] embed_docs = embeddings.embed_documents(split_docs_content) embed_query= embeddings.embed_query(query) from openai.embeddings_utils import cosine_similarity cos_index = [] for embed_doc in embed_docs: cos_index.append(cosine_similarity(embed_doc,embed_query)) #这边是根据大小建立索引 idx = sorted(range(len(cos_index)),key=lambda k:cos_index[k]) #根据cos_index的大小进行排序 final_similar_list = [] for index in idx[-3:]: unit = {} unit['sentences']=split_docs_content[index] unit['source']=split_docs[index]['source'] unit['score']=cos_index[index] final_similar_list.append(unit) return final_similar_list

2024-01-10

Kororinpas/Lit_Tool

embedding_function.py

def get_embedding_function(): from langchain.embeddings import HuggingFaceEmbeddings import torch device = 'cuda' if torch.cuda.is_available() else 'cpu' model_name = "sentence-transformers/all-mpnet-base-v2" model_kwargs = {'device':device} return HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs)

2024-01-10

Kororinpas/Lit_Tool

doi.py

def get_doi(abstract): from kor.extraction import create_extraction_chain from kor.nodes import Object, Text, Number from langchain.chat_models import ChatOpenAI llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0) # type: ignore schema = Object( id="doi", description="doi is a digital identifier.It typically starts with 10. followed by a numeric prefix, such as 10.1000/182.", attributes=[ Text( id="doi", description='doi is a digital identifier. It typically starts with "10." followed by a numeric prefix, such as 10.1000/182.', examples=[ ('American Economic Journal: Economic Policy 2015, 7(4): 223–242 http://dx.doi.org/10.1257/pol.20130367 223 Water Pollution Progress at Borders: The','http://dx.doi.org/10.1257/pol.20130367'), ('Environment and Development Economics (2020), 1–17 doi:10.1017/S1355770X2000025X EDE RESEARCH ARTICLE Political incentives, Party Congress, and pollution cycle: empirical evidence from China Zhihua Tian,1 and Yanfang Tian2* 1School of Economics, Zhejiang University of Technology, Hangzhou','10.1017/S1355770X2000025X') ], many=True ) ], many=False ) chain = create_extraction_chain(llm, schema, encoder_or_encoder_class='json') output = chain.predict_and_parse(text=abstract.page_content) if 'doi' not in output['data']: print(f"LLM strategy failed!!{abstract.metadata['source']} Please manually add it!!") source = 'None' return source else: doi = output['data']['doi']['doi'][0] if 'doi=' in doi: doi = doi.split('doi=')[1] return doi

2024-01-10

jied-O/Jids-Garage

langchainagentstest.py

from langchain import OpenAI from langchain.chains import LLMChain from langchain.chains import PALChain from langchain.agents import initialize_agent from langchain.agents import AgentType from langchain.agents import load_tools from ogbujipt.config import openai_emulation from ogbujipt.model_style.alpaca import prep_instru_inputs, ALPACA_PROMPT_TMPL from langchain.prompts import PromptTemplate openai_emulation(host="http://192.168.0.73", port="8000") def simpleWordPrompt(): prompt = PromptTemplate( input_variables=["place"], template="What is the capital of {place}?", ) print(prompt.format(place="Nigeria")) llm = OpenAI(temperature=0.1) llmchain = LLMChain(llm=llm, prompt=prompt) response = llmchain.run(place="Nigeria") print(response) def MathWorldProblem(): llm = OpenAI(temperature=0.1) palchain = PALChain.from_math_prompt(llm=llm, verbose=True) response = palchain.run( "If my age is half of my dad's age and he is going to be 60 next year, what is my current age?" ) print(response) def agentTest(): llm = OpenAI(temperature=0) tools = load_tools(["pal-math"], llm=llm) agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True) agent.run("If my age is half of my dad's age and he is going to be 60 next year, what is my current age?") def main(): MathWorldProblem() if __name__ == "__main__": main()

2024-01-10

tarunsamanta2k20/quivr

backend~parsers~audio.py

import os import tempfile import time from io import BytesIO from tempfile import NamedTemporaryFile import openai from fastapi import UploadFile from langchain.document_loaders import TextLoader from langchain.embeddings.openai import OpenAIEmbeddings from langchain.schema import Document from langchain.text_splitter import RecursiveCharacterTextSplitter from utils import compute_sha1_from_content, documents_vector_store # # Create a function to transcribe audio using Whisper # def _transcribe_audio(api_key, audio_file, stats_db): # openai.api_key = api_key # transcript = "" # with BytesIO(audio_file.read()) as audio_bytes: # # Get the extension of the uploaded file # file_extension = os.path.splitext(audio_file.name)[-1] # # Create a temporary file with the uploaded audio data and the correct extension # with tempfile.NamedTemporaryFile(delete=True, suffix=file_extension) as temp_audio_file: # temp_audio_file.write(audio_bytes.read()) # temp_audio_file.seek(0) # Move the file pointer to the beginning of the file # transcript = openai.Audio.translate("whisper-1", temp_audio_file) # return transcript async def process_audio(upload_file: UploadFile, stats_db): file_sha = "" dateshort = time.strftime("%Y%m%d-%H%M%S") file_meta_name = f"audiotranscript_{dateshort}.txt" # uploaded file to file object openai_api_key = os.environ.get("OPENAI_API_KEY") # Here, we're writing the uploaded file to a temporary file, so we can use it with your existing code. with tempfile.NamedTemporaryFile(delete=False, suffix=upload_file.filename) as tmp_file: await upload_file.seek(0) content = await upload_file.read() tmp_file.write(content) tmp_file.flush() tmp_file.close() with open(tmp_file.name, "rb") as audio_file: transcript = openai.Audio.transcribe("whisper-1", audio_file) file_sha = compute_sha1_from_content(transcript.text.encode("utf-8")) file_size = len(transcript.text.encode("utf-8")) # Load chunk size and overlap from sidebar chunk_size = 500 chunk_overlap = 0 text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder( chunk_size=chunk_size, chunk_overlap=chunk_overlap) texts = text_splitter.split_text(transcript) docs_with_metadata = [Document(page_content=text, metadata={"file_sha1": file_sha, "file_size": file_size, "file_name": file_meta_name, "chunk_size": chunk_size, "chunk_overlap": chunk_overlap, "date": dateshort}) for text in texts] # if st.secrets.self_hosted == "false": # add_usage(stats_db, "embedding", "audio", metadata={"file_name": file_meta_name,"file_type": ".txt", "chunk_size": chunk_size, "chunk_overlap": chunk_overlap}) documents_vector_store.add_documents(docs_with_metadata) return documents_vector_store

2024-01-10

sshh12/llm_optimize

llm_optimize~optimize.py

from typing import Callable, Optional, Tuple, List import re import openai from langchain.input import print_text from langchain.prompts.chat import ( SystemMessage, HumanMessage, AIMessage, ) from llm_optimize import llm, constants # The numeric score and the LLM-facing representation ScoreTuple = Tuple[float, str] # Best score, history of scores, best x0 OptimizationResultTuple = Tuple[float, List[float], str] def run( task_description: str, task_question: str, func: Callable[[str], ScoreTuple], x0: str, max_steps: Optional[int] = 10, model: Optional[llm.LLMModel] = None, verbose: Optional[bool] = True, system_prompt: Optional[str] = constants.SYSTEM_PROMPT, human_prompt: Optional[str] = constants.HUMAN_OPTIMIZATION_PROMPT, stop_score: Optional[float] = None, ) -> OptimizationResultTuple: if model is None: model = llm.get_default_llm() def _log(text: str, color: str): if verbose: print_text(text + "\n", color) x = x0 score, fx = func(x) best_score = score best_x = x _log(x, "blue") _log(fx, "green") messages = [ SystemMessage(content=system_prompt.format(task_description=task_description)), HumanMessage(content=human_prompt.format(task_question=task_question, x=x, fx=fx)), ] score_hist = [score] for _ in range(max_steps): try: resp = model(messages).content except openai.error.InvalidRequestError as e: _log(str(e), "red") # drop the first set of results to reduce token usage messages.pop(1) messages.pop(1) resp = model(messages).content _log(resp, "yellow") try: x = re.findall("```(?:\w+)?([\s\S]+)```", resp)[0] except IndexError as e: _log(f"Stopping early, failed to parse response. {e}", "red") break _log(x, "blue") score, fx = func(x) score_hist.append(score) if score > best_score: best_x = x best_score = score _log(fx, "green") messages.append(AIMessage(content=resp)) messages.append(HumanMessage(content=human_prompt.format(task_question=task_question, x=x, fx=fx))) if stop_score is not None and best_score >= stop_score: break return (best_score, score_hist, best_x)

2024-01-10

xiahan4956/Auto_Claude_100k

autogpt~llm~api_manager.py

from __future__ import annotations from typing import List, Optional import openai from openai import Model from autogpt.config import Config from autogpt.llm.base import CompletionModelInfo, MessageDict from autogpt.llm.providers.openai import OPEN_AI_MODELS from autogpt.logs import logger from autogpt.singleton import Singleton class ApiManager(metaclass=Singleton): def __init__(self): self.total_prompt_tokens = 0 self.total_completion_tokens = 0 self.total_cost = 0 self.total_budget = 0 self.models: Optional[list[Model]] = None def reset(self): self.total_prompt_tokens = 0 self.total_completion_tokens = 0 self.total_cost = 0 self.total_budget = 0.0 self.models = None def create_chat_completion( self, messages: list[MessageDict], model: str | None = None, temperature: float = None, max_tokens: int | None = None, deployment_id=None, ): """ Create a chat completion and update the cost. Args: messages (list): The list of messages to send to the API. model (str): The model to use for the API call. temperature (float): The temperature to use for the API call. max_tokens (int): The maximum number of tokens for the API call. Returns: str: The AI's response. """ cfg = Config() if temperature is None: temperature = cfg.temperature if deployment_id is not None: response = openai.ChatCompletion.create( deployment_id=deployment_id, model=model, messages=messages, temperature=temperature, max_tokens=max_tokens, api_key=cfg.openai_api_key, ) else: response = openai.ChatCompletion.create( model=model, messages=messages, temperature=temperature, max_tokens=max_tokens, api_key=cfg.openai_api_key, ) if not hasattr(response, "error"): logger.debug(f"Response: {response}") prompt_tokens = response.usage.prompt_tokens completion_tokens = response.usage.completion_tokens self.update_cost(prompt_tokens, completion_tokens, model) return response def update_cost(self, prompt_tokens, completion_tokens, model: str): """ Update the total cost, prompt tokens, and completion tokens. Args: prompt_tokens (int): The number of tokens used in the prompt. completion_tokens (int): The number of tokens used in the completion. model (str): The model used for the API call. """ # the .model property in API responses can contain version suffixes like -v2 model = model[:-3] if model.endswith("-v2") else model model_info = OPEN_AI_MODELS[model] self.total_prompt_tokens += prompt_tokens self.total_completion_tokens += completion_tokens self.total_cost += prompt_tokens * model_info.prompt_token_cost / 1000 if issubclass(type(model_info), CompletionModelInfo): self.total_cost += ( completion_tokens * model_info.completion_token_cost / 1000 ) logger.debug(f"Total running cost: ${self.total_cost:.3f}") def set_total_budget(self, total_budget): """ Sets the total user-defined budget for API calls. Args: total_budget (float): The total budget for API calls. """ self.total_budget = total_budget def get_total_prompt_tokens(self): """ Get the total number of prompt tokens. Returns: int: The total number of prompt tokens. """ return self.total_prompt_tokens def get_total_completion_tokens(self): """ Get the total number of completion tokens. Returns: int: The total number of completion tokens. """ return self.total_completion_tokens def get_total_cost(self): """ Get the total cost of API calls. Returns: float: The total cost of API calls. """ return self.total_cost def get_total_budget(self): """ Get the total user-defined budget for API calls. Returns: float: The total budget for API calls. """ return self.total_budget def get_models(self) -> List[Model]: """ Get list of available GPT models. Returns: list: List of available GPT models. """ if self.models is None: all_models = openai.Model.list()["data"] self.models = [model for model in all_models if "gpt" in model["id"]] return self.models

2024-01-10

xiahan4956/Auto_Claude_100k

autogpt~llm~utils~claude.py

from autogpt.config import Config import time import openai import json CFG = Config() openai.api_key = CFG.openai_api_key MAX_TOKEN_ONCE = 100000 CONTINUE_PROMPT = "... continue" from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT def _sendReq(anthropic, prompt, max_tokens_to_sample): print("----------------request----------------") print(prompt) print("----------------request----------------\n") print("the input words of claude: "+str(len(prompt))) for _ in range(5): try: response = anthropic.completions.create( prompt=prompt, stop_sequences = [HUMAN_PROMPT, AI_PROMPT], model="claude-2", max_tokens_to_sample=max_tokens_to_sample, temperature = 0.3 ) break except Exception as e: print(e) time.sleep(1) return response def sendReq(question, max_tokens_to_sample: int = MAX_TOKEN_ONCE): anthropic = Anthropic(api_key = CFG.claude_api_key) prompt = f"{question} {anthropic.AI_PROMPT}" response = _sendReq(anthropic, prompt, max_tokens_to_sample) data = response.completion return data def pmt_gpt_to_claude(question): question = str(question)[1:-1] question = question.replace("{\'role\': \'system\', \'content\':","\n\nSYSTEM:") question = question.replace("{\'role\': \'user\', \'content\':","\n\nHuman:") question = question.replace("{\'role\': \'assistant\', \'content\':","\n\nAssistant:") question = question.replace("\'}","") return question def fix_claude_json(claude_resp): messages = [{"role":"system","content":r"1. You will receive a JSON string, and your task is to extract information from it and return it as a JSON object. 2.Use function's json schema to extrct.Please notice the format 3. Be aware that the given JSON may contain errors, so you may need to infer the fields and the format from the JSON string. 4.Do not use \" and \' .you should use ' " },{"role": "user", "content": claude_resp}] functions = [ { "name": "parse_claude_json", "description": "parse a claude response to the json", "parameters": { "type": "object", "properties": { "thoughts": { "type": "object", "properties": { "text": { "type": "string", "description": "thoughts" }, "reasoning": { "type": "string" }, "plan": { "type": "string", "description": "it is a string,not list.If you find it is list,please use correct it " }, "criticism": { "type": "string", "description": "constructive self-criticism" }, "speak": { "type": "string", "description": "thoughts summary to say to user" } }, "required": ["text", "reasoning", "plan", "criticism", "speak"], }, "command": { "type": "object", "properties": { "name": {"type": "string"}, "args": { "type": "object" } }, "required": ["name", "args"], } }, "required": ["thoughts", "command"], }, }, ] resp_json = claude_resp for _ in range(5): try: response = openai.ChatCompletion.create( model="gpt-3.5-turbo-0613", messages=messages, functions=functions, max_tokens=3000, temperature=0.0, ) resp_json = response["choices"][0]["message"]["function_call"]["arguments"] break except Exception as e: time.sleep(1) print(e) # fix the plan try: resp_json = json.loads(resp_json) resp_json["thoughts"]["plan"] = str(resp_json["thoughts"]["plan"]).replace("[","").replace("]","") resp_json = json.dumps(resp_json) except Exception as e: print(e) return resp_json

2024-01-10

pkrack/asp

asp~ppo_patched.py

import warnings from typing import Any, Dict, Optional, Type, TypeVar, Union import numpy as np import torch as th from gymnasium import spaces from stable_baselines3.common.on_policy_algorithm import OnPolicyAlgorithm from stable_baselines3.common.policies import ActorCriticCnnPolicy, ActorCriticPolicy, BasePolicy, MultiInputActorCriticPolicy from stable_baselines3.common.type_aliases import GymEnv, MaybeCallback, Schedule from stable_baselines3.common.utils import explained_variance, get_schedule_fn from torch.nn import functional as F SelfPPO = TypeVar("SelfPPO", bound="PPO") class PPO(OnPolicyAlgorithm): """ Proximal Policy Optimization algorithm (PPO) (clip version) Paper: https://arxiv.org/abs/1707.06347 Code: This implementation borrows code from OpenAI Spinning Up (https://github.com/openai/spinningup/) https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail and Stable Baselines (PPO2 from https://github.com/hill-a/stable-baselines) Introduction to PPO: https://spinningup.openai.com/en/latest/algorithms/ppo.html :param policy: The policy model to use (MlpPolicy, CnnPolicy, ...) :param env: The environment to learn from (if registered in Gym, can be str) :param learning_rate: The learning rate, it can be a function of the current progress remaining (from 1 to 0) :param n_steps: The number of steps to run for each environment per update (i.e. rollout buffer size is n_steps * n_envs where n_envs is number of environment copies running in parallel) NOTE: n_steps * n_envs must be greater than 1 (because of the advantage normalization) See https://github.com/pytorch/pytorch/issues/29372 :param batch_size: Minibatch size :param n_epochs: Number of epoch when optimizing the surrogate loss :param gamma: Discount factor :param gae_lambda: Factor for trade-off of bias vs variance for Generalized Advantage Estimator :param clip_range: Clipping parameter, it can be a function of the current progress remaining (from 1 to 0). :param clip_range_vf: Clipping parameter for the value function, it can be a function of the current progress remaining (from 1 to 0). This is a parameter specific to the OpenAI implementation. If None is passed (default), no clipping will be done on the value function. IMPORTANT: this clipping depends on the reward scaling. :param normalize_advantage: Whether to normalize or not the advantage :param ent_coef: Entropy coefficient for the loss calculation :param vf_coef: Value function coefficient for the loss calculation :param max_grad_norm: The maximum value for the gradient clipping :param use_sde: Whether to use generalized State Dependent Exploration (gSDE) instead of action noise exploration (default: False) :param sde_sample_freq: Sample a new noise matrix every n steps when using gSDE Default: -1 (only sample at the beginning of the rollout) :param target_kl: Limit the KL divergence between updates, because the clipping is not enough to prevent large update see issue #213 (cf https://github.com/hill-a/stable-baselines/issues/213) By default, there is no limit on the kl div. :param tensorboard_log: the log location for tensorboard (if None, no logging) :param policy_kwargs: additional arguments to be passed to the policy on creation :param verbose: Verbosity level: 0 for no output, 1 for info messages (such as device or wrappers used), 2 for debug messages :param seed: Seed for the pseudo random generators :param device: Device (cpu, cuda, ...) on which the code should be run. Setting it to auto, the code will be run on the GPU if possible. :param _init_setup_model: Whether or not to build the network at the creation of the instance """ policy_aliases: Dict[str, Type[BasePolicy]] = { "MlpPolicy": ActorCriticPolicy, "CnnPolicy": ActorCriticCnnPolicy, "MultiInputPolicy": MultiInputActorCriticPolicy, } def __init__( self, policy: Union[str, Type[ActorCriticPolicy]], env: Union[GymEnv, str], learning_rate: Union[float, Schedule] = 3e-4, n_steps: int = 2048, batch_size: int = 64, n_epochs: int = 10, gamma: float = 0.99, gae_lambda: float = 0.95, clip_range: Union[float, Schedule] = 0.2, clip_range_vf: Union[None, float, Schedule] = None, normalize_advantage: bool = True, ent_coef: float = 0.0, vf_coef: float = 0.5, max_grad_norm: float = 0.5, use_sde: bool = False, sde_sample_freq: int = -1, target_kl: Optional[float] = None, tensorboard_log: Optional[str] = None, policy_kwargs: Optional[Dict[str, Any]] = None, verbose: int = 0, seed: Optional[int] = None, device: Union[th.device, str] = "auto", _init_setup_model: bool = True, ): super().__init__( policy, env, learning_rate=learning_rate, n_steps=n_steps, gamma=gamma, gae_lambda=gae_lambda, ent_coef=ent_coef, vf_coef=vf_coef, max_grad_norm=max_grad_norm, use_sde=use_sde, sde_sample_freq=sde_sample_freq, tensorboard_log=tensorboard_log, policy_kwargs=policy_kwargs, verbose=verbose, device=device, seed=seed, _init_setup_model=False, supported_action_spaces=( spaces.Box, spaces.Discrete, spaces.MultiDiscrete, spaces.MultiBinary, ), ) # Sanity check, otherwise it will lead to noisy gradient and NaN # because of the advantage normalization if normalize_advantage: assert ( batch_size > 1 ), "`batch_size` must be greater than 1. See https://github.com/DLR-RM/stable-baselines3/issues/440" if self.env is not None: # Check that `n_steps * n_envs > 1` to avoid NaN # when doing advantage normalization buffer_size = self.env.num_envs * self.n_steps assert buffer_size > 1 or ( not normalize_advantage ), f"`n_steps * n_envs` must be greater than 1. Currently n_steps={self.n_steps} and n_envs={self.env.num_envs}" # Check that the rollout buffer size is a multiple of the mini-batch size untruncated_batches = buffer_size // batch_size if buffer_size % batch_size > 0: warnings.warn( f"You have specified a mini-batch size of {batch_size}," f" but because the `RolloutBuffer` is of size `n_steps * n_envs = {buffer_size}`," f" after every {untruncated_batches} untruncated mini-batches," f" there will be a truncated mini-batch of size {buffer_size % batch_size}\n" f"We recommend using a `batch_size` that is a factor of `n_steps * n_envs`.\n" f"Info: (n_steps={self.n_steps} and n_envs={self.env.num_envs})" ) self.batch_size = batch_size self.n_epochs = n_epochs self.clip_range = clip_range self.clip_range_vf = clip_range_vf self.normalize_advantage = normalize_advantage self.target_kl = target_kl if _init_setup_model: self._setup_model() def _setup_model(self) -> None: super()._setup_model() # Initialize schedules for policy/value clipping self.clip_range = get_schedule_fn(self.clip_range) if self.clip_range_vf is not None: if isinstance(self.clip_range_vf, (float, int)): assert self.clip_range_vf > 0, "`clip_range_vf` must be positive, " "pass `None` to deactivate vf clipping" self.clip_range_vf = get_schedule_fn(self.clip_range_vf) def train(self) -> None: """ Update policy using the currently gathered rollout buffer. """ # Switch to train mode (this affects batch norm / dropout) self.policy.set_training_mode(True) # Update optimizer learning rate self._update_learning_rate(self.policy.optimizer) # Compute current clip range clip_range = self.clip_range(self._current_progress_remaining) # Optional: clip range for the value function if self.clip_range_vf is not None: clip_range_vf = self.clip_range_vf(self._current_progress_remaining) entropy_losses = [] pg_losses, value_losses = [], [] clip_fractions = [] continue_training = True loss = None for epoch in range(self.n_epochs): approx_kl_divs = [] # Do a complete pass on the rollout buffer for rollout_data, bc_data in self.rollout_buffer.get(self.batch_size): # Re-sample the noise matrix because the log_std has changed if self.use_sde: self.policy.reset_noise(self.batch_size) if bc_data is None: bc_loss = th.zeros(1, device=self.device) else: _, log_probs, _ = self.policy.evaluate_actions(bc_data.obs, bc_data.action) ratio = th.exp(log_probs - bc_data.log_prob) bc_loss = -th.mean(th.clamp(ratio, 1 - clip_range, 1 + clip_range)).to(self.device) if rollout_data is not None: actions = rollout_data.actions if isinstance(self.action_space, spaces.Discrete): # Convert discrete action from float to long actions = rollout_data.actions.long().flatten() values, log_prob, entropy = self.policy.evaluate_actions(rollout_data.observations, actions) values = values.flatten() # Normalize advantage advantages = rollout_data.advantages # Normalization does not make sense if mini batchsize == 1, see GH issue #325 if self.normalize_advantage and len(advantages) > 1: advantages = (advantages - advantages.mean()) / (advantages.std() + 1e-8) # ratio between old and new policy, should be one at the first iteration ratio = th.exp(log_prob - rollout_data.old_log_prob) # clipped surrogate loss policy_loss_1 = advantages * ratio policy_loss_2 = advantages * th.clamp(ratio, 1 - clip_range, 1 + clip_range) policy_loss = -th.min(policy_loss_1, policy_loss_2).mean() # Logging pg_losses.append(policy_loss.item()) clip_fraction = th.mean((th.abs(ratio - 1) > clip_range).float()).item() clip_fractions.append(clip_fraction) if self.clip_range_vf is None: # No clipping values_pred = values else: # Clip the difference between old and new value # NOTE: this depends on the reward scaling values_pred = rollout_data.old_values + th.clamp( values - rollout_data.old_values, -clip_range_vf, clip_range_vf ) # Value loss using the TD(gae_lambda) target value_loss = F.mse_loss(rollout_data.returns, values_pred) value_losses.append(value_loss.item()) # Entropy loss favor exploration if entropy is None: # Approximate entropy when no analytical form entropy_loss = -th.mean(-log_prob) else: entropy_loss = -th.mean(entropy) entropy_losses.append(entropy_loss.item()) # Calculate approximate form of reverse KL Divergence for early stopping # see issue #417: https://github.com/DLR-RM/stable-baselines3/issues/417 # and discussion in PR #419: https://github.com/DLR-RM/stable-baselines3/pull/419 # and Schulman blog: http://joschu.net/blog/kl-approx.html with th.no_grad(): log_ratio = log_prob - rollout_data.old_log_prob approx_kl_div = th.mean((th.exp(log_ratio) - 1) - log_ratio).cpu().numpy() approx_kl_divs.append(approx_kl_div) if self.target_kl is not None and approx_kl_div > 1.5 * self.target_kl: continue_training = False if self.verbose >= 1: print(f"Early stopping at step {epoch} due to reaching max kl: {approx_kl_div:.2f}") break loss = (policy_loss + self.ent_coef * entropy_loss + self.vf_coef * value_loss) + self.bc_coef * bc_loss else: loss = bc_loss # Optimization step self.policy.optimizer.zero_grad() loss.backward() # Clip grad norm th.nn.utils.clip_grad_norm_(self.policy.parameters(), self.max_grad_norm) self.policy.optimizer.step() if not continue_training: break if loss: self._n_updates += self.n_epochs explained_var = explained_variance(self.rollout_buffer.values.flatten(), self.rollout_buffer.returns.flatten()) # Logs if loss: self.logger.record("train/entropy_loss", np.mean(entropy_losses)) self.logger.record("train/policy_gradient_loss", np.mean(pg_losses)) self.logger.record("train/value_loss", np.mean(value_losses)) self.logger.record("train/approx_kl", np.mean(approx_kl_divs)) self.logger.record("train/clip_fraction", np.mean(clip_fractions)) self.logger.record("train/loss", loss.item()) self.logger.record("train/explained_variance", explained_var) if hasattr(self.policy, "log_std"): self.logger.record("train/std", th.exp(self.policy.log_std).mean().item()) self.logger.record("train/n_updates", self._n_updates, exclude="tensorboard") self.logger.record("train/clip_range", clip_range) if self.clip_range_vf is not None: self.logger.record("train/clip_range_vf", clip_range_vf) else: self.logger.info("No valid goals in the batch, skipping update") def learn( self: SelfPPO, total_timesteps: int, callback: MaybeCallback = None, log_interval: int = 1, tb_log_name: str = "PPO", reset_num_timesteps: bool = True, progress_bar: bool = False, ) -> SelfPPO: return super().learn( total_timesteps=total_timesteps, callback=callback, log_interval=log_interval, tb_log_name=tb_log_name, reset_num_timesteps=reset_num_timesteps, progress_bar=progress_bar, )

2024-01-10

jongio/chat-with-your-data-solution-accelerator

backend~utilities~orchestrator~Strategies.py

from enum import Enum class OrchestrationStrategy(Enum): OPENAI_FUNCTION = 'openai_function' LANGCHAIN = 'langchain' def get_orchestrator(orchestration_strategy: str): if orchestration_strategy == OrchestrationStrategy.OPENAI_FUNCTION.value: from .OpenAIFunctions import OpenAIFunctionsOrchestrator return OpenAIFunctionsOrchestrator() elif orchestration_strategy == OrchestrationStrategy.LANGCHAIN.value: from .LangChainAgent import LangChainAgent return LangChainAgent() else: raise Exception(f"Unknown orchestration strategy: {orchestration_strategy}")

2024-01-10

jongio/chat-with-your-data-solution-accelerator

backend~utilities~document_chunking~Layout.py

from typing import List from .DocumentChunkingBase import DocumentChunkingBase from langchain.text_splitter import MarkdownTextSplitter from .Strategies import ChunkingSettings from ..common.SourceDocument import SourceDocument class LayoutDocumentChunking(DocumentChunkingBase): def __init__(self) -> None: pass def chunk(self, documents: List[SourceDocument], chunking: ChunkingSettings) -> List[SourceDocument]: full_document_content = "".join(list(map(lambda document: document.content, documents))) document_url = documents[0].source splitter = MarkdownTextSplitter.from_tiktoken_encoder(chunk_size=chunking.chunk_size, chunk_overlap=chunking.chunk_overlap) chunked_content_list = splitter.split_text(full_document_content) # Create document for each chunk documents = [] chunk_offset = 0 for idx, chunked_content in enumerate(chunked_content_list): documents.append( SourceDocument.from_metadata( content=chunked_content, document_url=document_url, metadata={"offset": chunk_offset}, idx=idx, ) ) chunk_offset += len(chunked_content) return documents

2024-01-10

jongio/chat-with-your-data-solution-accelerator

backend~utilities~helpers~LLMHelper.py

import openai from typing import List from langchain.chat_models import AzureChatOpenAI from langchain.embeddings import OpenAIEmbeddings from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler from .EnvHelper import EnvHelper class LLMHelper: def __init__(self): env_helper: EnvHelper = EnvHelper() # Configure OpenAI API openai.api_type = "azure" openai.api_version = env_helper.AZURE_OPENAI_API_VERSION openai.api_base = env_helper.OPENAI_API_BASE openai.api_key = env_helper.OPENAI_API_KEY self.llm_model = env_helper.AZURE_OPENAI_MODEL self.llm_max_tokens = env_helper.AZURE_OPENAI_MAX_TOKENS if env_helper.AZURE_OPENAI_MAX_TOKENS != '' else None self.embedding_model = env_helper.AZURE_OPENAI_EMBEDDING_MODEL def get_llm(self): return AzureChatOpenAI(deployment_name=self.llm_model, temperature=0, max_tokens=self.llm_max_tokens, openai_api_version=openai.api_version) # TODO: This needs to have a custom callback to stream back to the UI def get_streaming_llm(self): return AzureChatOpenAI(streaming=True, callbacks=[StreamingStdOutCallbackHandler], deployment_name=self.llm_model, temperature=0, max_tokens=self.llm_max_tokens, openai_api_version=openai.api_version) def get_embedding_model(self): return OpenAIEmbeddings(deployment=self.embedding_model, chunk_size=1) def get_chat_completion_with_functions(self, messages: List[dict], functions: List[dict], function_call: str="auto"): return openai.ChatCompletion.create( deployment_id=self.llm_model, messages=messages, functions=functions, function_call=function_call, ) def get_chat_completion(self, messages: List[dict]): return openai.ChatCompletion.create( deployment_id=self.llm_model, messages=messages, )

2024-01-10

pcc2k00/HousingPriceTrend

HousingPriceTrendMetaphor.py

import openai import yaml from metaphor_python import Metaphor with open("pass.yml") as f: content = f.read() my_credentials = yaml.load(content, Loader=yaml.FullLoader) openai.api_key = my_credentials["openAi"] metaphor = Metaphor(my_credentials["metaphor"]) USER_QUESTION = "Recent housing price in Seattle" SYSTEM_MESSAGE = "You are a helpful assistant that generates search queiries based on user questions. Only generate one search query." completion = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": SYSTEM_MESSAGE}, {"role": "user", "content": USER_QUESTION}, ], ) query = completion.choices[0].message.content search_response = metaphor.search( query, use_autoprompt=True, start_published_date="2023-07-01" ) contents_result = search_response.get_contents() first_result = contents_result.contents[0] SYSTEM_MESSAGE = "You are a helpful assistant that summarizes the content of a webpage. Summarize the users input." completion = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": SYSTEM_MESSAGE}, {"role": "user", "content": first_result.extract}, ], ) summary = completion.choices[0].message.content print(f"Summary for {first_result.title}: {summary}")

2024-01-10

romain-cambonie/openxcom-mod-generator

src~chat~ask_for_visual_proposition.py

from openai import OpenAI from openai.types.chat import ChatCompletion def ask_for_concept_art( client: OpenAI, character_story: str, art_style_description: str, ) -> str: system_prompt = ( "Generate a comprehensive and vivid visual concept art of a character for a piece of artwork. " "The character should fit within a distinct theme and style, and the description must be detailed enough to guide an " "artist in creating a dynamic and engaging image." "Here are the guidelines for your description:" "Theme and Setting: Choose an intriguing theme and setting for the character. It could be anything from a dystopian " "future to a fantasy world. " "Describe the setting in a way that complements the character's story and personality." "Character Details:" "Physical Appearance: Provide a detailed description of the character's physical features, including hair, eyes, " "skin, and build." "Expression and Posture: Convey the character's mood or personality through their expression and posture." "Attire and Equipment: Describe the character's clothing and any distinctive equipment they might carry, " "do NOT use proper noun, describe visually what the items look like." f"Artistic Style: Specify the desired artistic style for the portrayal. The starting point is : " f"{art_style_description}, make sure to detail the stylistic elements that should be emphasized." "Composition and Color Palette: Suggest a striking composition for the artwork" "Describe the character stance" "Describe the color palette, considering how colors can reflect the character's traits or the mood of the setting." "Extract up to 8 keys focusing on the art style and composition" "Use these guidelines to create a structured and detailed visual description for a character based on the following " "origin story:" "Focus on making the description as vivid and detailed as possible, so it can easily be translated into a stunning " "piece of art." "" "An example of a good concept art result:" "Keys: Commanding presence, Dynamic composition, Low angle perspective, Cold metallic shades, Warm leather tones, " "Dramatic lighting, Cyberpunk aesthetic" "Character Details: She is light-skinned with a muscular build, short blonde hair, and piercing light-colored eyes " "that radiate intelligence and cunning. Her expression is one of chilling neutrality, a reflection of her spirit " "shaped by the cold, ruthless Arctic." "Attire and Equipment: Her attire combines functionality with a touch of brutality – a sleek, black chest armor that " "bulges with the strength of her physique, complemented by large shoulder pads. Her arms are covered with highly " "detailed armor, and her legs are clad in thigh-high boots with sturdy knee pads. Fortified gloves adorn her hands. " "In one hand, she deftly holds a leather whip, an emblem of elegance and cruelty, while her other hand grips a robust " "submachine gun. Around her waist are vials containing clear liquid and spherical objects reminiscent of primitive " "grenades, adding to her enigmatic persona. A handle and a battle axe, symbols of her defiance and skill, " "are fastened at her side." "Setting: The backdrop is a post-apocalyptic Arctic tundra, subtly hinting at her origins. The environment should be " "bleak yet captivating, with remnants of a once-thriving world now lost to chaos and rebellion." "Artistic Style and Composition: The portrait should capture her commanding presence amidst this desolate backdrop. " "The composition should be dynamic, focusing on her from a slightly low angle to emphasize her dominance. The color " "palette should be a blend of cold metallic shades and warmer tones from her leather armor, creating a vivid contrast " "that underscores her determination and grit. The lighting should be dramatic, highlighting her features and the " "textures of her gear, enhancing the overall cyberpunk aesthetic." ) user_prompt = f"Character story: {character_story}" response: ChatCompletion = client.chat.completions.create( model="gpt-4", messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}], ) return str(response.choices[0].message.content)

2024-01-10

romain-cambonie/openxcom-mod-generator

src~dalle~call_dalle_and_save_image.py

import requests from openai import OpenAI from pathlib import Path from typing import Optional from openai.types import ImagesResponse def call_dalle_and_save_image(prompt: str, client: OpenAI, output_file_path: Path) -> Optional[Path]: try: # Generate image using OpenAI client response: ImagesResponse = client.images.generate( prompt=prompt, n=1, model="dall-e-3", size="1024x1024", quality="hd", response_format="url" ) # Extract the image URL image_url = response.data[0].url if not image_url: print("No image URL found in the response.") return None print(image_url) # Download the image image_response = requests.get(image_url) if image_response.status_code == 200: # Write the image data to a file with open(output_file_path, "wb") as file: file.write(image_response.content) return output_file_path else: print(f"Error downloading image: {image_response.status_code}") return None except Exception as e: print(f"An error occurred: {e}") return None

2024-01-10

romain-cambonie/openxcom-mod-generator

src~chat~ask_for_dalle_character_prompt.py

from openai import OpenAI from openai.types.chat import ChatCompletion def ask_for_dalle_character_prompt( client: OpenAI, concept_art_description: str, ) -> str: system_prompt = ( "You're given a detailed concept art description of a character. Your task is to condense this description into a " "succinct, vivid DALL-E prompt." "The DALL-E prompt should accurately capture the key visual elements and artistic style described in the concept art, " "while being concise enough for effective image generation. " "Here is the concept art description to be transformed into a DALL-E prompt:\n" f"{concept_art_description}\n" "Based on this description, refine this concept into a DALL-E prompt that contains, in order references to the art " "style, composition, subject, location, colors;" "The prompt must not be more than 130 words, encapsulating the essence of the concept art." f"The prompt must start with the keys of the concept art" ) user_prompt = "Transform the above concept art description into a succinct DALL-E prompt." response: ChatCompletion = client.chat.completions.create( model="gpt-4", messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}], ) return str(response.choices[0].message.content)

2024-01-10

romain-cambonie/openxcom-mod-generator

src~chat~ask_for_origin_story.py

from openai import OpenAI from openai.types.chat import ChatCompletion def ask_for_origin_story( client: OpenAI, character_name: str, equipment_description: str, appearance_description: str, ) -> str: system_prompt = ( "You are tasked with creating a short origin story for a fictional character. " "You will receive three key pieces of information: (1) the character's name, " "(2) a YAML payload detailing the character's equipment, and " "(3) an image that shows some characteristics of the character's appearance. " "Your job is to weave these elements together into a compelling and imaginative origin story. " "The story should be concise, no more than a few paragraphs, and should creatively incorporate specific details from " "the YAML payload and the visual cues from the image. " "The tone and style of the story should align with the genre suggested by the character's name and appearance. " "Be imaginative and ensure that the equipment and visual traits play a significant role in the character's background " "and the events that shaped them." "Pay special attention to match all visual description details such as gender, race, skin color, hair color and so on " ) user_prompt = ( f"Character Name: {character_name}\n\nEquipment: {equipment_description}\n\nAppearance: " f"{appearance_description}\n\nBased on the above details, create a short origin story for the character." ) response: ChatCompletion = client.chat.completions.create( model="gpt-4", messages=[{"role": "system", "content": system_prompt}, {"role": "user", "content": user_prompt}], ) return str(response.choices[0].message.content)

2024-01-10

outlines-dev/outlines

outlines~models~__init__.py

"""Module that contains all the models integrated in outlines. We group the models in submodules by provider instead of theme (completion, chat completion, diffusers, etc.) and use routing functions everywhere else in the codebase. """ from .awq import awq from .exllamav2 import exl2 from .gptq import gptq from .llamacpp import LlamaCpp, llamacpp from .mamba import Mamba, mamba from .openai import OpenAI, openai from .transformers import Transformer, transformers

2024-01-10

ball2004244/Pinecone-Hackathon-23-Backend

logic~pinecone_db.py

''' This file contains the logic for storing and querying data from Pinecone. ''' from typing import List from langchain.vectorstores import Pinecone from langchain.chains.summarize import load_summarize_chain from langchain.llms import GooglePalm from langchain.embeddings.google_palm import GooglePalmEmbeddings from langchain.schema import Document import pinecone from pinecone import DescribeIndexStatsResponse class PineconeTrainer: def __init__(self, gcp_api_key: str, pinecone_api_key: str, pinecone_environment: str): self.gcp_api_key = gcp_api_key self.pinecone_api_key = pinecone_api_key self.pinecone_environment = pinecone_environment self.palm_config = { 'temperature': 0.7, 'google_api_key': self.gcp_api_key, } self.index_name = 'paragraph-summarizer' self.llm = GooglePalm(**self.palm_config) self.chain = load_summarize_chain(self.llm, chain_type='stuff') self.embeddings = GooglePalmEmbeddings(**self.palm_config) self.pinecone_init(self.index_name, 'cosine', 768) def pinecone_init(self, index_name: str, metric: str, dimension: int) -> None: pinecone.init( api_key=self.pinecone_api_key, environment=self.pinecone_environment, ) # check if index exists if index_name not in pinecone.list_indexes(): pinecone.create_index(name=index_name, metric=metric, dimension=dimension) self.index = pinecone.Index(index_name=index_name) self.vectordb = Pinecone(index=self.index, embedding_function=self.embeddings.embed_query, text_key='text') def add_data(self, input_list: List[str]=[]) -> None: document_list = [Document(page_content=input_list[i]) for i in range(len(input_list))] self.vectordb = Pinecone.from_documents(document_list, embedding=self.embeddings, index_name=self.index_name) print('Data added successfully!, %s vectors added' % len(input_list)) def delete_all_data(self) -> None: pass def query(self, query: str=' ', question: str='Summarize in 3 sentences') -> str: search = self.vectordb.similarity_search(query=query, k=3) summary = self.chain.run(input_documents=search, question=question) return summary def get_index_info(self) -> DescribeIndexStatsResponse: index = pinecone.GRPCIndex(self.index_name) output = index.describe_index_stats() return output def embed_text(self, text: str) -> List[float]: return self.embeddings.embed_query(text) def pinecone_train(self, input_file: str) -> None: try: input_list = self.extract_input_text(input_file) self.add_data(input_list) except Exception as e: print(e) @staticmethod def extract_input_text(input_file: str) -> List[str]: from logic.data_extract import extract_data, extract_text data = extract_data(input_file) texts = extract_text(data) return texts @staticmethod def extract_output_text(input_file: str) -> List[str]: from logic.data_extract import extract_data, extract_output_text data = extract_data(input_file) texts = extract_output_text(data) return texts if __name__ == '__main__': pass

2024-01-10

TheoKanning/crossword

crossword~clues.py

import json import os import openai def convert_raw_clues(raw_filename, output_filename): """ Reads raw clue info from raw_filename, formats it to match GPT-3's fine-tune input, and writes it to output_filename Raw clues are formatted like "Up in the air : ALOFT" """ with open(output_filename, "w+") as f_out: f_out.write("farts") with open(raw_filename, "r") as f_in: with open(output_filename, "w+") as f_out: for line in f_in.readlines(): line = line.strip() if not line: continue if line.isnumeric(): # This line is a clue number, ignore it continue if line.lower() == "down" or line.lower() == "across": continue components = line.rsplit( ":", 1 ) # split from end to handle colons inside clues if len(components) != 2: print(line) continue clue = components[0].strip() answer = components[1].strip() f_out.write( json.dumps( { "prompt": f"Answer: {answer.lower()}\nClue:", "completion": f" {clue}\n", } ) ) f_out.write("\n") def get_clue(answer): prompt = f"Answer: {answer.lower()}\nClue:" openai.api_key = os.getenv("OPENAI_API_KEY") result = openai.Completion.create( model="curie:ft-personal-2022-04-30-18-38-57", prompt=prompt, stop="\n", n=5 ) print(f"Answer: {answer}\nClues:") for choice in result["choices"]: print(choice["text"]) if __name__ == "__main__": get_clue("") # convert_raw_clues("../clues/raw_clues.txt", "../clues/formatted.jsonl")

2024-01-10

NusretOzates/langchain_retrieval_qa_bot

data_loaders.py

import re from itertools import chain from typing import List from langchain.docstore.document import Document from langchain.document_loaders import PyPDFLoader, TextLoader, UnstructuredURLLoader from langchain.indexes import VectorstoreIndexCreator from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.vectorstores import DocArrayInMemorySearch from langchain.vectorstores.base import VectorStoreRetriever def load_text_file(file_path: str) -> Document: """Loads a text file and returns a Document object. Args: file_path: Path to the text file. Returns: A Document object. """ doc = TextLoader(file_path, encoding="utf-8").load()[0] return doc def load_pdf_file(file_path: str) -> List[Document]: """Loads a pdf file and returns a list of Document objects. Args: file_path: Path to the pdf file. Returns: A list of Document objects. Every page in the pdf file is a Document object. """ loader = PyPDFLoader(file_path) docs = loader.load() return docs def load_website(url: str) -> List[Document]: """Loads a website and returns a Document object. Args: url: Url of the website. Returns: A Document object. """ documents = UnstructuredURLLoader( [url], mode="elements", headers={ "ssl_verify": "False", }, ).load() processed_docs = [] # We are not rich, we need to eliminate some of the elements for doc in documents: # This will make us lose table information sorry about that :( if doc.metadata.get("category") not in [ "NarrativeText", "UncategorizedText", "Title", ]: continue # Remove elements with empty links, they are mostly recommended articles etc. if doc.metadata.get("links"): link = doc.metadata["links"][0]["text"] if link is None: continue link = link.replace(" ", "").replace("\n", "") if len(link.split()) == 0: continue # Remove titles with links, they are mostly table of contents or navigation links if doc.metadata.get("category") == "Title" and doc.metadata.get("links"): continue # Remove extra spaces doc.page_content = re.sub(" +", " ", doc.page_content) # Remove docs with less than 3 words if len(doc.page_content.split()) < 3: continue processed_docs.append(doc) # Instead of splitting element-wise, we merge all the elements and split them in chunks merged_docs = "\n".join([doc.page_content for doc in processed_docs]) splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100) processed_docs = splitter.split_text(merged_docs) processed_docs = [ Document(page_content=doc, metadata={"url": url}) for doc in processed_docs ] return processed_docs def load_text_files(file_paths: List[str]) -> List[Document]: """Loads a list of text files and returns a list of Document objects. Args: file_paths: List of paths to the text files. Returns: A list of Document objects. """ docs = [load_text_file(file_path) for file_path in file_paths] return docs def load_pdf_files(file_paths: List[str]) -> List[Document]: """Loads a list of pdf files and returns a list of Document objects. Args: file_paths: List of paths to the pdf files. Returns: A list of Document objects. Every page in the pdf file is a Document object. """ docs = [load_pdf_file(file_path) for file_path in file_paths] docs = list(chain.from_iterable(docs)) return docs def create_index(docs: List[Document]) -> VectorStoreRetriever: """Creates a vectorstore index from a list of Document objects. Args: docs: List of Document objects. Returns: A vectorstore index. It searches the most similar document to the given query but with the help of MMR it also tries to find the most diverse document to the given query. """ index = VectorstoreIndexCreator( vectorstore_cls=DocArrayInMemorySearch, text_splitter=RecursiveCharacterTextSplitter( chunk_size=1000, chunk_overlap=100 ), ).from_documents(docs) return index.vectorstore.as_retriever(search_type="mmr")

2024-01-10

Antrozhuk/telegramChatGPTBot

src~telegram_bot.py

import telegram.constants as constants from telegram import Update from telegram.ext import ApplicationBuilder, ContextTypes, CommandHandler, MessageHandler, filters from src.openai_helper import OpenAIHelper from src.logger import Logger class ChatGPT3TelegramBot: """ Class representing a Chat-GPT3 Telegram Bot. """ def __init__(self, config: dict, openai: OpenAIHelper): """ Ініціалізує бот конфігурацією та GPT-3 налаштуваннями. :param config: Словник з конфігурацією бота :param openai: OpenAIHelper обʼєкт :param disallowed_message: Повідомлення про відсутність доступу """ self.config = config self.openai = openai self.logger = Logger('telegram_bot').get_logger() self.disallowed_message = "Вибачте, але вам не дозволено користуватись цим ботом." async def start(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Показує початкове повідомлення. """ if await self.disallowed(update, context): return await update.message.reply_text("Привіт! Я бот, який відповідає на ваші повідомлення за допомогою ChatGPT-3.\n" "Якщо ви хочете дізнатись більше про мене, введіть /help\n\n", disable_web_page_preview=True) async def help(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Показує допоміжне повідомлення. """ if await self.disallowed(update, context): return await update.message.reply_text("[Будь яке повідомлення] - Відправляє ваше повідомлення до AI\n" "/help - Меню помічника\n" "/random_answer - Генерує рандомну відповідь\n" "/random_post - Генерує рандомний пост\n" "/reset - Оновлює бесіду\n\n", disable_web_page_preview=True) async def reset(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Оновлює бесіду. """ if await self.disallowed(update, context): return self.logger.info(f'Resetting the conversation for {update.message.from_user}...') chat_id = update.effective_chat.id self.openai.reset_chat_history(chat_id=chat_id) await context.bot.send_message(chat_id=chat_id, text='Готово!') async def prompt(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ React to incoming messages and respond accordingly. """ if await self.disallowed(update, context): return self.logger.info(f'New message "{update.message.text}" received from {update.message.from_user}') chat_id = update.effective_chat.id await context.bot.send_chat_action(chat_id=chat_id, action=constants.ChatAction.TYPING) response = self.openai.get_chat_response(chat_id=chat_id, query=update.message.text) await context.bot.send_message( chat_id=chat_id, reply_to_message_id=update.message.id, parse_mode=constants.ParseMode.MARKDOWN, text=response ) async def random_answer(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Відправляє рандомну відповідь. """ if await self.disallowed(update, context): return self.logger.info(f'random_answer command received from {update.message.from_user}') chat_id = update.effective_chat.id await context.bot.send_chat_action(chat_id=chat_id, action=constants.ChatAction.TYPING) response = self.openai.get_chat_response(chat_id=chat_id, query='напиши рандомну відповідь') await context.bot.send_message( chat_id=chat_id, reply_to_message_id=update.message.id, parse_mode=constants.ParseMode.MARKDOWN, text=response ) async def random_post(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Відправляє рандомний пост. """ if await self.disallowed(update, context): return self.logger.info(f'random_post command received from {update.message.from_user}') chat_id = update.effective_chat.id await context.bot.send_chat_action(chat_id=chat_id, action=constants.ChatAction.TYPING) response = self.openai.get_chat_response(chat_id=chat_id, query='напиши рандомний пост українською') await context.bot.send_message( chat_id=chat_id, parse_mode=constants.ParseMode.MARKDOWN, text=response ) async def disallowed(self, update: Update, context: ContextTypes.DEFAULT_TYPE): """ Відправляє повідомлення про відсутність доступів до користувача. """ if not await self.is_allowed(update): self.logger.warning(f'User {update.message.from_user} is not allowed to use the bot') await context.bot.send_message( chat_id=update.effective_chat.id, text=self.disallowed_message, disable_web_page_preview=True ) return True return False async def error_handler(self, update: object, context: ContextTypes.DEFAULT_TYPE) -> None: """ Відловлює всі помилки. """ self.logger.debug(f'Exception while handling an update: {context.error}') async def is_allowed(self, update: Update) -> bool: """ Перевіряє чи дозволено юзеру користуватись даним ботом. """ if self.config['allowed_user_ids'] == '*': return True allowed_user_ids = self.config['allowed_user_ids'].split(',') if str(update.message.from_user.id) in allowed_user_ids: return True return False def run(self): """ Запускає бот доки користувач не натисне Ctrl+C """ application = ApplicationBuilder().token(self.config['token']).build() application.add_handler(CommandHandler('start', self.start)) application.add_handler(CommandHandler('help', self.help)) application.add_handler(CommandHandler('reset', self.reset)) application.add_handler(CommandHandler('random_answer', self.random_answer)) application.add_handler(CommandHandler('random_post', self.random_post)) application.add_handler(MessageHandler(filters.TEXT & (~filters.COMMAND), self.prompt)) application.add_error_handler(self.error_handler) application.run_polling()

2024-01-10

aws-samples/aurora-postgresql-pgvector

DAT303~02_QuestionAndAnswering~rag_app.py

# Import libraries from dotenv import load_dotenv from PyPDF2 import PdfReader from langchain.vectorstores.pgvector import PGVector from langchain.memory import ConversationSummaryBufferMemory from langchain.chains import ConversationalRetrievalChain from htmlTemplates import css from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.embeddings import BedrockEmbeddings from langchain.llms import Bedrock from langchain.prompts import PromptTemplate import streamlit as st import boto3 from PIL import Image import os import traceback # TODO: This function takes a list of PDF documents as input and extracts the text from them using PdfReader. # It concatenates the extracted text and returns it. # TODO: Given the extracted text, this function splits it into smaller chunks using the RecursiveCharacterTextSplitter module. # The chunk size, overlap, and other parameters are configured to optimize processing efficiency. # TODO: This function takes the text chunks as input and creates a vector store using Bedrock Embeddings (Titan) and pgvector. # The vector store stores the vector representations of the text chunks, enabling efficient retrieval based on semantic similarity. # TODO: In this function, a conversation chain is created using the conversational AI model (Anthropic's Claude v2), vector store (created in the previous function), and conversation memory (ConversationSummaryBufferMemory). # This chain allows the Gen AI app to engage in conversational interactions. # This function is responsible for processing the user's input question and generating a response from the chatbot def handle_userinput(user_question): if "chat_history" not in st.session_state: st.session_state.chat_history = None if "messages" not in st.session_state: st.session_state.messages = [] try: response = st.session_state.conversation({'question': user_question}) except ValueError: st.write("Sorry, I didn't understand that. Could you rephrase your question?") print(traceback.format_exc()) return st.session_state.chat_history = response['chat_history'] for i, message in enumerate(st.session_state.chat_history): if i % 2 == 0: st.success(message.content, icon="🤔") else: st.write(message.content) # Streamlit components def main(): # Set the page configuration for the Streamlit application, including the page title and icon. st.set_page_config(page_title="Generative AI Q&A with Amazon Bedrock, Aurora PostgreSQL and pgvector", layout="wide", page_icon=":books::parrot:") st.write(css, unsafe_allow_html=True) logo_url = "static/Powered-By_logo-stack_RGB_REV.png" st.sidebar.image(logo_url, width=150) st.sidebar.markdown( """ ### Instructions: 1. Browse and upload PDF files 2. Click Process 3. Type your question in the search bar to get more insights """ ) # Check if the conversation and chat history are not present in the session state and initialize them to None. if "conversation" not in st.session_state: st.session_state.conversation = get_conversation_chain(get_vectorstore(None)) if "chat_history" not in st.session_state: st.session_state.chat_history = None # A header with the text appears at the top of the Streamlit application. st.header("Generative AI Q&A with Amazon Bedrock, Aurora PostgreSQL and pgvector :books::parrot:") subheader = '<p style="font-family:Calibri (Body); color:Grey; font-size: 16px;">Leverage Foundational Models from <a href="https://aws.amazon.com/bedrock/">Amazon Bedrock</a> and <a href="https://github.com/pgvector/pgvector">pgvector</a> as Vector Engine</p>' # Write the CSS style to the Streamlit application, allowing you to customize the appearance. st.markdown(subheader, unsafe_allow_html=True) image = Image.open("static/RAG_APG.png") st.image(image, caption='Generative AI Q&A with Amazon Bedrock, Aurora PostgreSQL and pgvector') # Create a text input box where you can ask questions about your documents. user_question = st.text_input("Ask a question about your documents:", placeholder="What is Amazon Aurora?") # Define a Go button for user action go_button = st.button("Submit", type="secondary") # If the go button is pressed or the user enters a question, it calls the handle_userinput() function to process the user's input. if go_button or user_question: with st.spinner("Processing..."): handle_userinput(user_question) with st.sidebar: st.subheader("Your documents") pdf_docs = st.file_uploader( "Upload your PDFs here and click on 'Process'", type="pdf", accept_multiple_files=True) # If the user clicks the "Process" button, the following code is executed: # i. raw_text = get_pdf_text(pdf_docs): retrieves the text content from the uploaded PDF documents. # ii. text_chunks = get_text_chunks(raw_text): splits the text content into smaller chunks for efficient processing. # iii. vectorstore = get_vectorstore(text_chunks): creates a vector store that stores the vector representations of the text chunks. if st.button("Process"): with st.spinner("Processing"): # get pdf text raw_text = get_pdf_text(pdf_docs) # get the text chunks text_chunks = get_text_chunks(raw_text) # create vector store vectorstore = get_vectorstore(text_chunks) # create conversation chain st.session_state.conversation = get_conversation_chain(vectorstore) st.success('PDF uploaded successfully!', icon="✅") with st.sidebar: st.divider() st.sidebar.markdown( """ ### Sample questions to get started: 1. What is Amazon Aurora? 2. How can I migrate from PostgreSQL to Aurora and the other way around? 3. What does "three times the performance of PostgreSQL" mean? 4. What is Aurora Standard and Aurora I/O-Optimized? 5. How do I scale the compute resources associated with my Amazon Aurora DB Instance? 6. How does Amazon Aurora improve my databases fault tolerance to disk failures? 7. How does Aurora improve recovery time after a database crash? 8. How can I improve upon the availability of a single Amazon Aurora database? """ ) if __name__ == '__main__': # This function loads the environment variables from a .env file. load_dotenv() # Define the Bedrock client. BEDROCK_CLIENT = boto3.client("bedrock-runtime", 'us-west-2') # Create the connection string for pgvector from .env file. CONNECTION_STRING = PGVector.connection_string_from_db_params( driver = os.environ.get("PGVECTOR_DRIVER"), user = os.environ.get("PGVECTOR_USER"), password = os.environ.get("PGVECTOR_PASSWORD"), host = os.environ.get("PGVECTOR_HOST"), port = os.environ.get("PGVECTOR_PORT"), database = os.environ.get("PGVECTOR_DATABASE") ) main()

2024-01-10

WuQingYi20/InteractiveStory

wsgi.py

from flask import Flask, render_template, jsonify, request import openai import re from prompts import prompts from dotenv import load_dotenv import os # Load the .env file load_dotenv() app = Flask(__name__) initialCall = True currentDescription = "" # Initialize OpenAI API with your API key openai.api_key = os.getenv('OPENAI_API_KEY') # Define a dictionary to store user progress data user_data = {} # Global variable to track initialization status initialized = False @app.route('/') def index(): global initialized global currentDescription if initialized: # Initialization has already been done, return JSON response if request.headers.get('X-Requested-With') == 'XMLHttpRequest': return jsonify(story=user_data['story'], choices=user_data['choices']) # Initialization has already been done, return HTML response else: return render_template('index.html', story=user_data['story'], choices=user_data['choices']) else: # Initialization code systemRoleAuto = prompts['index']['System'] promptStory = prompts['index']['story'] storyResponse = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": f"{systemRoleAuto}"}, {"role": "user", "content": f"{promptStory}"}, #{"role": "assistant", "content": f"{contentAssistant}"}, ], max_tokens= 1500, ) story = storyResponse.choices[0].message['content'] currentDescription = story choicesPrompt = prompts['index']['choices'] choiceResponse = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": f"{systemRoleAuto}"}, {"role": "user", "content": f"{story} {choicesPrompt}"}, #{"role": "assistant", "content": f"{contentAssistant}"}, ], max_tokens= 1500, ) #Insert <p> tags around each paragraph formatted_story = format_story(story) user_data['story'] = formatted_story user_data['choices'] = choiceResponse.choices[0].message['content'] initialized = True if request.headers.get('X-Requested-With') == 'XMLHttpRequest': return jsonify(story=story, choices=user_data['choices']) else: return render_template('index.html', story=story, choices=user_data['choices']) # Define a route to handle user choices and update the story @app.route('/next-page/<choice>') def next_page(choice): systemRoleAuto = prompts['next-page']['System'] originalStory = user_data['story'] + "\n" + choice contentAssistant = prompts['next-page']['storyAssistant'] contentAssistantChoices = prompts['next-page']['choicesAssistant'] prompt_story = originalStory + "\n" + prompts['next-page']['story'] response_story = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": f"{systemRoleAuto}"}, {"role": "user", "content": f"{prompt_story}"}, {"role": "assistant", "content": f"{contentAssistant}"}, ], max_tokens= 1500, ) prompt_choices = originalStory + response_story.choices[0].message['content'] + "\n" + prompts['next-page']['choices'] response_choices = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": f"{systemRoleAuto}"}, {"role": "user", "content": f"{prompt_choices}"}, {"role": "assistant", "content": f"{contentAssistantChoices}"}, ], max_tokens= 1500, ) story = response_story.choices[0].message['content'] choices = response_choices.choices[0].message['content'] # get summary of previous story and actions by gpt-3.5-turbo and original story prompt_summary = originalStory + "\n" + prompts['next-page']['summary'] response_summary = openai.ChatCompletion.create( model="gpt-3.5-turbo", messages=[ {"role": "system", "content": f"{systemRoleAuto}"}, {"role": "user", "content": f"{prompt_summary}"}, #{"role": "assistant", "content": f"{contentAssistant}"}, ], max_tokens= 1500, ) formatted_story = format_story(story) user_data['story'] = formatted_story user_data['choices'] = choices user_data['summary'] = response_summary.choices[0].message['content'] return jsonify(story=formatted_story, choices=choices, summary=user_data['summary']) def format_story(story): # Split the text into paragraphs using a regular expression paragraphs = re.split(r"\n\s*\n", story) #Insert <p> tags around each paragraph formatted_story = "\n".join([f"<p>{paragraph}</p>" for paragraph in paragraphs]) return formatted_story if __name__ == '__main__': app.run(debug=True)

2024-01-10

yamdereneko/ymbot

src~chatGPT~Chat_GPT_API.py

# -*- coding: utf-8 -*- import asyncio import nonebot from pydantic import BaseModel from httpx import AsyncClient import src.Data.jx3_Redis as redis import openai class Response(BaseModel): """返回数据模型""" id: str """状态码""" object: str created: int model: str choices: list """返回消息字符串""" usage: dict | list[dict] """返回数据""" class ChatGPTAPI: client: AsyncClient def __init__(self): proxy_url = "http://username:password@127.0.0.1:8888" proxies = {"http": proxy_url, "https": proxy_url} self.client = AsyncClient(proxies=proxies) self.url = "https://api.openai.com/v1/completions" async def call_api(self, content) -> Response: red = redis.Redis() chat_gpt_apikey = await red.query("chat_gpt_apikey") Organization = await red.query("OpenAI-Organization") """请求api网站数据""" headers = { 'Authorization': f'Bearer {chat_gpt_apikey}', 'OpenAI-Organization': Organization, 'Content-Type': 'application/json' } data = { "model": "gpt-3.5-turbo", "messages": [{"role": "user", "content": content}] } res = await self.client.post(url=self.url, json=data, headers=headers, timeout=3000) print(res) nonebot.logger.info(res.text) return Response.parse_obj(res.json())