mirror of
https://github.com/nomic-ai/gpt4all
synced 2024-11-08 07:10:32 +00:00
c4706d0c14
* porting over replit code model to gpt4all * replaced memory with kv_self struct * continuing debug * welp it built but lot of sus things * working model loading and somewhat working generate.. need to format response? * revert back to semi working version * finally got rid of weird formatting * figured out problem is with python bindings - this is good to go for testing * addressing PR feedback * output refactor * fixed prompt reponse collection * cleanup * addressing PR comments * building replit backend with new ggmlver code * chatllm replit and clean python files * cleanup * updated replit to match new llmodel api * match llmodel api and change size_t to Token * resolve PR comments * replit model commit comment
114 lines
3.1 KiB
Python
114 lines
3.1 KiB
Python
from pathlib import Path
|
|
import sys
|
|
import struct
|
|
import json
|
|
import numpy as np
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
import sentencepiece.sentencepiece_model_pb2 as model
|
|
|
|
if len(sys.argv) < 3:
|
|
print("Usage: convert-h5-to-ggml.py dir-model [use-f32]\n")
|
|
print(" ftype == 0 -> float32")
|
|
print(" ftype == 1 -> float16")
|
|
sys.exit(1)
|
|
|
|
|
|
# output in the same directory as the model
|
|
dir_model = sys.argv[1]
|
|
fname_out = sys.argv[1] + "/ggml-replit-code-v1-3b.bin"
|
|
|
|
|
|
with open(dir_model + "/config.json", "r", encoding="utf-8") as f:
|
|
hparams = json.load(f)
|
|
|
|
sp_proto = model.ModelProto()
|
|
sp_proto.ParseFromString(open(Path(sys.argv[1]) / "spiece.model", "rb").read())
|
|
|
|
|
|
# possible data types
|
|
# ftype == 0 -> float32
|
|
# ftype == 1 -> float16
|
|
#
|
|
# map from ftype to string
|
|
ftype_str = ["f32", "f16"]
|
|
|
|
ftype = 1
|
|
if len(sys.argv) > 2:
|
|
ftype = int(sys.argv[2])
|
|
if ftype < 0 or ftype > 1:
|
|
print("Invalid ftype: " + str(ftype))
|
|
sys.exit(1)
|
|
fname_out = sys.argv[1] + "/ggml-replit-code-v1-3b-" + ftype_str[ftype] + ".bin"
|
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(dir_model, trust_remote_code=True)
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
dir_model, low_cpu_mem_usage=True, trust_remote_code=True
|
|
)
|
|
# print (model)
|
|
|
|
# print(tokenizer.encode('I believe the meaning of life is'))
|
|
|
|
list_vars = model.state_dict()
|
|
for name in list_vars.keys():
|
|
print(name, list_vars[name].shape, list_vars[name].dtype)
|
|
|
|
fout = open(fname_out, "wb")
|
|
|
|
print(hparams)
|
|
|
|
fout.write(struct.pack("i", 0x7265706c)) # magic: repl in hex
|
|
fout.write(struct.pack("i", hparams["vocab_size"]))
|
|
fout.write(struct.pack("i", hparams["max_seq_len"]))
|
|
fout.write(struct.pack("i", hparams["d_model"]))
|
|
fout.write(struct.pack("i", hparams["n_heads"]))
|
|
fout.write(struct.pack("i", hparams["n_layers"]))
|
|
fout.write(struct.pack("i", ftype))
|
|
|
|
|
|
# TODO: temporary hack to not deal with implementing the tokenizer
|
|
for piece in sp_proto.pieces:
|
|
encoded_piece = piece.piece.encode("utf-8")
|
|
fout.write(struct.pack("i", len(encoded_piece)))
|
|
fout.write(encoded_piece)
|
|
fout.write(struct.pack("f", piece.score))
|
|
|
|
|
|
for name in list_vars.keys():
|
|
data = list_vars[name].squeeze().numpy()
|
|
print("Processing variable: " + name + " with shape: ", data.shape)
|
|
|
|
n_dims = len(data.shape)
|
|
|
|
# ftype == 0 -> float32, ftype == 1 -> float16
|
|
ftype_cur = 0
|
|
if ftype != 0:
|
|
if name[-7:] == ".weight" and n_dims == 2:
|
|
print(" Converting to float16")
|
|
data = data.astype(np.float16)
|
|
ftype_cur = 1
|
|
else:
|
|
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)
|
|
|
|
fout.close()
|
|
|
|
print("Done. Output file: " + fname_out)
|
|
print("")
|