zhangyiss/mlx-examples
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx-examples.git synced 2025-06-27 03:05:20 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

chore: make the Deepseek example compatible with Yi models. (#205)

Browse Source 
* Update convert.py

* Update convert.py

* Update deepseek_coder.py
This commit is contained in:
Anchen 2023-12-31 01:11:33 +11:00 committed by GitHub
parent 581a5733a1
commit ee3c44d231
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
llms/deepseek-coder/convert.py
View File

@ -44,7 +44,7 @@ def convert(args):
config = model.config.to_dict() config = model.config.to_dict()
state_dict = model.state_dict() state_dict = model.state_dict()
tokenizer = AutoTokenizer.from_pretrained(str(hf_path), trust_remote_code=True) tokenizer = AutoTokenizer.from_pretrained(str(hf_path), trust_remote_code=True, use_fast=False)
# things to change # things to change
# 1. there's no "model." in the weight names # 1. there's no "model." in the weight names
@ -84,7 +84,7 @@ def convert(args):
weights = {k: v.numpy() for k, v in state_dict.items()} weights = {k: v.numpy() for k, v in state_dict.items()}
config["rope_scaling_factor"] = config["rope_scaling"]["factor"] config["rope_scaling_factor"] = config["rope_scaling"]["factor"] if config["rope_scaling"] is not None else 1.0
keep_keys = set( keep_keys = set(
[ [
"vocab_size", "vocab_size",
@ -96,6 +96,7 @@ def convert(args):
"rms_norm_eps", "rms_norm_eps",
"intermediate_size", "intermediate_size",
"rope_scaling_factor", "rope_scaling_factor",
"rope_theta"
] ]
) )
for k in list(config.keys()): for k in list(config.keys()):

2
llms/deepseek-coder/deepseek_coder.py
View File

@ -248,7 +248,7 @@ def load_model(model_path: str):
nn.QuantizedLinear.quantize_module(model, **quantization) nn.QuantizedLinear.quantize_module(model, **quantization)
model.update(tree_unflatten(list(weights.items()))) model.update(tree_unflatten(list(weights.items())))
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True) tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, use_fast=False)
return model, tokenizer return model, tokenizer