initial commit: model files

README.md

@@ -0,0 +1,57 @@
+---
+tags:
+- conversational
+- tagalog
+- filipino
+
+language:
+- tl 
+
+inference: false
+---
+
+# Tagalog DialoGPT
+
+A DialoGPT model fine-tuned on Tagalog conversational data scraped from the web. This model is an output of a research on BERT-based data augmentation for low resource languages. We fine-tuned DialoGPT-medium as our base model.
+
+#  Latest release: July 25, 2021
+* As of the moment, the model is only able to respond based on the history of 3 previous utterances before being limited. This is a result of the scarce amount of Tagalog conversations in our dataset.
+
+
+# Dataset and Scripts
+The training data used was collected under the following categories:
+* Food and Drinks
+* Home and Garden
+* Style and Fashion
+* Travel and Leisure
+* Visas and Immigration
+* Health and Wellness
+* Body and Fitness
+* Small Talk
+
+Pinoy Exchange (PEx) Conversational Dataset to be released soon.
+
+# Usage
+
+Here is an example of using Beam Search as the decoding method for our model.
+```
+for step in range(2):
+    # encode the new user input, add the eos_token and return a tensor in Pytorch
+    new_user_input_ids = tokenizer.encode(input(">> User:") + tokenizer.eos_token, return_tensors='pt')
+
+    # append the new user input tokens to the chat history
+    bot_input_ids = torch.cat([chat_history_ids, new_user_input_ids], dim=-1) if step > 0 else new_user_input_ids
+
+    # we limit the generation to 512 tokens, each utterance in training had a maximum of 128 tokens
+    chat_history_ids = model.generate(
+        bot_input_ids, max_length=512,
+        pad_token_id=tokenizer.eos_token_id,
+        num_beams=5, 
+        no_repeat_ngram_size=3
+    )
+    
+    # pretty print last ouput tokens from bot
+    print("DialoGPT: {}".format(tokenizer.decode(chat_history_ids[:, bot_input_ids.shape[-1]:][0], skip_special_tokens=True)))
+```
+
+

config.json

@@ -0,0 +1,30 @@
+{
+  "activation_function": "gelu_new",
+  "architectures": [
+    "GPT2LMHeadModel"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50256,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 1024,
+  "n_head": 16,
+  "n_layer": 24,
+  "n_positions": 1024,
+  "resid_pdrop": 0.1,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "conversational": {
+      "max_length": 1000
+    }
+  },
+  "vocab_size": 50257
+}

eval_results.txt

@@ -0,0 +1 @@
+perplexity = tensor(4.1206)

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f9b235d5f4062b2104bebedbbf8de1410a3615460f040f7f98ac5d62336fa53
+size 1444581337

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<|endoftext|>", "eos_token": "<|endoftext|>", "unk_token": "<|endoftext|>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4780f61cd642ebfbd3838e556ce3bc5525498cb9607cbf0f78d865ced16dac8c
+size 1327