2023-05-06 15:37:25 -04:00
|
|
|
# Convert Hugging Face fine-tuned bloom-like models to ggml format
|
|
|
|
#
|
|
|
|
# Usage:
|
|
|
|
#
|
|
|
|
# python3 models/convert-h5-to-ggml.py
|
|
|
|
#
|
|
|
|
# This script is similar to "convert-pt-to-ggml.py"
|
|
|
|
#
|
|
|
|
|
|
|
|
import io
|
|
|
|
import os
|
|
|
|
import sys
|
|
|
|
import struct
|
|
|
|
import json
|
|
|
|
import code
|
|
|
|
import torch
|
|
|
|
import numpy as np
|
|
|
|
|
|
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoConfig, BloomForCausalLM
|
|
|
|
|
|
|
|
# ref: https://github.com/openai/gpt-2/blob/master/src/encoder.py
|
|
|
|
def bytes_to_unicode():
|
|
|
|
"""
|
|
|
|
Returns list of utf-8 byte and a corresponding list of unicode strings.
|
|
|
|
The reversible bpe codes work on unicode strings.
|
|
|
|
This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
|
|
|
|
When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
|
|
|
|
This is a significant percentage of your normal, say, 32K bpe vocab.
|
|
|
|
To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
|
|
|
|
And avoids mapping to whitespace/control characters the bpe code barfs on.
|
|
|
|
"""
|
|
|
|
bs = list(range(ord("!"), ord("~")+1))+list(range(ord("¡"), ord("¬")+1))+list(range(ord("®"), ord("ÿ")+1))
|
|
|
|
cs = bs[:]
|
|
|
|
n = 0
|
|
|
|
for b in range(2**8):
|
|
|
|
if b not in bs:
|
|
|
|
bs.append(b)
|
|
|
|
cs.append(2**8+n)
|
|
|
|
n += 1
|
|
|
|
cs = [chr(n) for n in cs]
|
|
|
|
return dict(zip(bs, cs))
|
|
|
|
|
|
|
|
if len(sys.argv) < 3:
|
|
|
|
print("Usage: python convert-hf-to-ggml.py model_name dir-output [use-f32]")
|
|
|
|
print(" model_name: name of the model to convert. Example: 'bigscience/bloomz-560m'")
|
|
|
|
print(" dir-output: directory where the output file will be written")
|
|
|
|
print(" use-f32: if present, use float32 instead of float16")
|
|
|
|
sys.exit(1)
|
|
|
|
|
|
|
|
model_name = sys.argv[1]
|
|
|
|
dir_out = sys.argv[2]
|
|
|
|
|
|
|
|
# make sure the output directory exists
|
|
|
|
os.makedirs(dir_out, exist_ok=True)
|
|
|
|
|
|
|
|
# possible data types
|
|
|
|
# ftype == 0 -> float32
|
|
|
|
# ftype == 1 -> float16
|
|
|
|
#
|
|
|
|
# map from ftype to string
|
|
|
|
ftype_str = ["f32", "f16"]
|
|
|
|
ftype = 1
|
|
|
|
if len(sys.argv) > 3:
|
|
|
|
ftype = 0
|
|
|
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
|
|
|
config = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
|
|
|
|
hparams = config.to_dict()
|
|
|
|
print("Loading model: ", model_name)
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, config=config, torch_dtype=torch.float16 if ftype == 1 else torch.float32, low_cpu_mem_usage=True)
|
|
|
|
print("Model loaded: ", model_name)
|
|
|
|
|
|
|
|
|
|
|
|
fname_out = dir_out + f"/ggml-model-{model_name.split('/')[-1]}-{ftype_str[ftype]}.bin"
|
|
|
|
fout = open(fname_out, "wb")
|
2023-05-07 18:18:04 -04:00
|
|
|
vocab = tokenizer.vocab
|
2023-05-06 15:37:25 -04:00
|
|
|
|
|
|
|
hparams["multiple_of"] = 1
|
2023-05-11 11:20:43 -04:00
|
|
|
fout.write(struct.pack("I", 0x67676d6d)) # magic: ggml in hex
|
|
|
|
fout.write(struct.pack("I", model.config.vocab_size))
|
|
|
|
fout.write(struct.pack("I", model.config.max_seq_len))
|
|
|
|
fout.write(struct.pack("I", model.config.n_layers))
|
|
|
|
fout.write(struct.pack("I", model.config.n_heads))
|
|
|
|
fout.write(struct.pack("I", model.config.d_model))
|
|
|
|
fout.write(struct.pack("f", model.config.attn_config['alibi_bias_max']))
|
|
|
|
clip_qkv = model.config.attn_config['clip_qkv']
|
|
|
|
fout.write(struct.pack("f", clip_qkv if clip_qkv is not None else 0))
|
|
|
|
fout.write(struct.pack("I", ftype))
|
2023-05-06 15:37:25 -04:00
|
|
|
|
|
|
|
# # Is this correct??
|
|
|
|
# dot_token = tokenizer.encode(".")[0]
|
2023-05-07 18:18:04 -04:00
|
|
|
# write tokens to ggml file
|
2023-05-11 11:20:43 -04:00
|
|
|
dot_token = tokenizer.encode('.')[0]
|
|
|
|
fout.write(struct.pack("I", model.config.vocab_size))
|
|
|
|
|
|
|
|
for i in range(model.config.vocab_size):
|
|
|
|
text = tokenizer.decode([dot_token, i]).encode('utf-8')
|
|
|
|
# remove the first byte (it's always '.')
|
|
|
|
text = text[1:]
|
|
|
|
enclen = len(text)
|
|
|
|
if i in tokenizer.all_special_ids:
|
|
|
|
print(f"special token: {text}")
|
|
|
|
enclen = enclen | 1<<31
|
|
|
|
fout.write(struct.pack("I", enclen))
|
2023-05-07 18:18:04 -04:00
|
|
|
fout.write(text)
|
2023-05-06 15:37:25 -04:00
|
|
|
|
|
|
|
list_vars = model.state_dict()
|
|
|
|
for name in list_vars.keys():
|
|
|
|
data = list_vars[name].squeeze().numpy()
|
|
|
|
print("Processing variable: " + name + " with shape: ", data.shape)
|
|
|
|
|
2023-05-11 11:20:43 -04:00
|
|
|
n_dims = len(data.shape);
|
2023-05-06 15:37:25 -04:00
|
|
|
|
2023-05-11 11:20:43 -04:00
|
|
|
# ftype == 0 -> float32, ftype == 1 -> float16
|
|
|
|
ftype_cur = 0;
|
|
|
|
if ftype != 0:
|
|
|
|
# Keep token embeddings in fp32
|
|
|
|
if name[-7:] == ".weight" and n_dims == 2 and ".wte" not in name:
|
|
|
|
print(" Converting to float16")
|
|
|
|
data = data.astype(np.float16)
|
|
|
|
ftype_cur = 1
|
2023-05-06 15:37:25 -04:00
|
|
|
else:
|
2023-05-11 11:20:43 -04:00
|
|
|
print(" Converting to float32")
|
|
|
|
data = data.astype(np.float32)
|
|
|
|
ftype_cur = 0
|
|
|
|
else:
|
|
|
|
if data.dtype != np.float32:
|
|
|
|
print(" Converting to float32")
|
|
|
|
data = data.astype(np.float32)
|
|
|
|
ftype_cur = 0
|
|
|
|
|
|
|
|
# header
|
|
|
|
str = name.encode('utf-8')
|
|
|
|
fout.write(struct.pack("iii", n_dims, len(str), ftype_cur))
|
|
|
|
for i in range(n_dims):
|
|
|
|
fout.write(struct.pack("i", data.shape[n_dims - 1 - i]))
|
|
|
|
fout.write(str);
|
|
|
|
|
|
|
|
# data
|
|
|
|
data.tofile(fout)
|
2023-05-06 15:37:25 -04:00
|
|
|
|
|
|
|
fout.close()
|
|
|
|
|
|
|
|
print("Done. Output file: " + fname_out)
|
|
|
|
print("")
|