delete file

btlm_3b_convert.py

@@ -1,242 +0,0 @@
-# -*- coding: utf-8 -*-
-"""btlm_3b_convert.ipynb
-
-Automatically generated by Colaboratory.
-
-Original file is located at
-    https://colab.research.google.com/drive/1kuWDvfRlc0BplrityIIwpdf404mljhRK
-"""
-
-!pip install -U transformers
-!pip install -U accelerate #git+https://github.com/huggingface/accelerate.git
-!pip install -U bitsandbytes #git+ https://github.com/timdettmers/bitsandbytes.git
-
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-# Load the tokenizer and model
-tokenizer = AutoTokenizer.from_pretrained("cerebras/btlm-3b-8k-base")
-model = AutoModelForCausalLM.from_pretrained(
-    "cerebras/btlm-3b-8k-base",
-    trust_remote_code=True,
-    torch_dtype="auto",
-    load_in_8bit=True,
-    offload_folder="offload",
-  )
-
-# Set the prompt for generating text
-prompt = "Albert Einstein was known for "
-
-# Tokenize the prompt and convert to PyTorch tensors
-inputs = tokenizer(prompt, return_tensors="pt")
-
-# Generate text using the model
-outputs = model.generate(
-    **inputs,
-    num_beams=5,
-    max_new_tokens=50,
-    early_stopping=True,
-    no_repeat_ngram_size=2
-)
-
-# Convert the generated token IDs back to text
-generated_text = tokenizer.batch_decode(outputs, skip_special_tokens=True)
-
-# Print the generated text
-print(generated_text[0])
-
-print(model)
-
-list_vars = model.state_dict()
-for name in list_vars.keys():
-    print(name, "=>", list_vars[name])
-
-import sys
-import os
-import struct
-import json
-
-import torch
-from transformers import AutoConfig
-
-config = AutoConfig.from_pretrained("cerebras/btlm-3b-8k-base", trust_remote_code=True)
-hparams = config.to_dict()
-fname_out = "btlm-3b.ggml.bin"
-
-print(json.dumps(hparams, indent=4, sort_keys=True))
-
-import re
-import numpy as np
-import tensorflow as tf
-
-bfloat16 = tf.bfloat16.as_numpy_dtype
-
-
-fout = open(fname_out, "wb")
-
-fout.write(struct.pack("i", 0x67676D6C))
-fout.write(struct.pack("i", hparams["vocab_size"]))
-fout.write(struct.pack("i", hparams["n_positions"]))
-fout.write(struct.pack("i", hparams["n_embd"]))
-fout.write(struct.pack("i", hparams["n_head"]))
-fout.write(struct.pack("i", hparams["n_layer"]))
-fout.write(struct.pack("i", hparams["n_inner"]))
-fout.write(struct.pack("i", 1))
-
-for i in range(hparams["vocab_size"]):
-  text = tokenizer.decode([i]).encode('utf-8')
-  fout.write(struct.pack("i", len(text)))
-  fout.write(text)
-
-
-# for name in list_vars.keys():
-#     print(name, "=>", list_vars[name])
-
-
-for name in list_vars.keys():
-    if name[-14:] == ".weight_format":
-      print("FOUND " + name)
-      continue
-
-
-
-
-    print("Processing variable: " + name)
-    data = list_vars[name].squeeze().cpu().type(dtype=torch.float16).numpy()
-    print(" with shape: ", data.shape)
-
-    # rename headers to keep compatibility
-    if name == "transformer.ln_f.weight":
-        name = "model/ln_f/g"
-    elif name == "transformer.ln_f.bias":
-        name = "model/ln_f/b"
-    elif name == "transformer.wte.weight":
-        name = "model/wte"
-    elif name == "transformer.wpe.weight":
-        name = "model/wpe"
-    elif name == "lm_head.weight":
-        name = "model/lm_head"
-    elif name == "transformer.relative_pe.slopes":
-      name = "model/relative_pe/slopes"
-    elif re.match(r"transformer.h\.\d+\.ln_1\.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/ln_1/g"
-    elif re.match(r"transformer.h\.\d+\.ln_1\.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/ln_1/b"
-    elif re.match(r"transformer.h\.\d+\.attn\.c_attn\.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_attn/w"
-    elif re.match(r"transformer.h\.\d+\.attn\.c_attn\.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_attn/b"
-    elif re.match(r"transformer.h\.\d+\.attn\.c_proj\.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_proj/w"
-    elif re.match(r"transformer.h.\d+.attn.c_proj.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_proj/b"
-    elif re.match(r"transformer.h.\d+.ln_2.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/ln_2/g"
-    elif re.match(r"transformer.h.\d+.ln_2.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/ln_2/b"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc/w"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc/b"
-    elif re.match(r"transformer.h.\d+.mlp.c_proj.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_proj/w"
-    elif re.match(r"transformer.h.\d+.mlp.c_proj.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_proj/b"
-    # NEW
-    elif re.match(r"transformer.h.\d+.attn.c_proj.SCB", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_proj/scb"
-    elif re.match(r"transformer.h.\d+.attn.c_attn.SCB", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/attn/c_attn/scb"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc.SCB", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc/scb"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc2.weight", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc2/weight"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc2.bias", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc2/bias"
-    elif re.match(r"transformer.h.\d+.mlp.c_fc2.SCB", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_fc2/scb"
-    elif re.match(r"transformer.h.\d+.mlp.c_proj.SCB", name):
-        i = re.findall("\d+", name)[0]
-        name = f"model/h{i}/mlp/c_proj/scb"
-
-    else:
-        print("Unrecognized variable name. %s", name)
-
-
-    # we don't need these
-    if name.endswith("attn.masked_bias") or name.endswith(".attn.bias"):
-        print("  Skipping variable: " + name)
-        continue
-
-    n_dims = len(data.shape);
-
-    # ftype == 0 -> float32, ftype == 1 -> float16
-    ftype = 0;
-    if True: #use_b16
-        if (name == "model/wte" or name == "model/lm_head" or name[-2:] == "/g" or name[-2:] == "/w") and n_dims == 2:
-            print("  Converting to float16")
-            data = data.astype(np.float16)
-            ftype = 1
-        else:
-            print("  Converting to float32")
-            data = data.astype(np.float32)
-            ftype = 0
-
-    # for efficiency - transpose the projection matrices
-    # "model/h.*/attn/c_attn/w"
-    # "model/h.*/attn/c_proj/w"
-    # "model/h.*/mlp/c_fc/w"
-    # "model/h.*/mlp/c_proj/w"
-    if name[-14:] == "/attn/c_attn/w" or \
-       name[-14:] == "/attn/c_proj/w" or \
-       name[-11:] == "/mlp/c_fc/w" or \
-       name[-13:] == "/mlp/c_proj/w":
-        print("  Transposing")
-        data = data.transpose()
-
-    # header
-    str = name.encode('utf-8')
-    fout.write(struct.pack("iii", n_dims, len(str), ftype))
-    for i in range(n_dims):
-        fout.write(struct.pack("i", data.shape[n_dims - 1 - i]))
-    fout.write(str);
-
-    # data
-    data.tofile(fout)
-
-fout.close()
-
-print("Done. Output file: " + fname_out)
-print("")
-
-
-# write_binary()
-
-from huggingface_hub import login, HfApi
-
-login()
-
-api = HfApi()
-
-api.upload_folder(
-    folder_path="/content/btlm-3b-ggml",
-    repo_id="bornjre/btlm-3b-ggml",
-    repo_type="model",
-)