mirror/OpenLLM
1
0
Fork 0
mirror of https://github.com/bentoml/OpenLLM.git synced 2026-02-18 22:55:08 -05:00
Code Issues Packages Projects Releases Wiki Activity

chore(style): reduce line length and truncate compression

Browse Source 
Signed-off-by: aarnphm-ec2-dev <29749331+aarnphm@users.noreply.github.com>
This commit is contained in:
aarnphm-ec2-dev
2023-08-22 17:02:00 +00:00
parent bc851b1d13
commit eddbc06374
67 changed files with 1661 additions and 301 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
openllm-python/src/openllm/models/starcoder/modeling_starcoder.py
View File

@@ -27,7 +27,12 @@ class StarCoder(openllm.LLM["transformers.GPTBigCodeForCausalLM", "transformers.
with torch.inference_mode():
# eos_token_id=self.tokenizer.convert_tokens_to_ids("<|end|>"), # NOTE: this is for finetuning starcoder
# NOTE: support fine-tuning starcoder
result_tensor = self.model.generate(self.tokenizer.encode(prompt, return_tensors="pt").to(self.device), do_sample=True, pad_token_id=self.tokenizer.eos_token_id, generation_config=self.config.model_construct_env(**attrs).to_generation_config())
result_tensor = self.model.generate(
self.tokenizer.encode(prompt, return_tensors="pt").to(self.device),
do_sample=True,
pad_token_id=self.tokenizer.eos_token_id,
generation_config=self.config.model_construct_env(**attrs).to_generation_config()
)
# TODO: We will probably want to return the tokenizer here so that we can manually process this
# return (skip_special_tokens=False, clean_up_tokenization_spaces=False))
return self.tokenizer.batch_decode(result_tensor[0], skip_special_tokens=True, clean_up_tokenization_spaces=True)