Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Added possibility to generate base text on GPU for text evaluation. #1945

Merged
merged 3 commits into from
Mar 21, 2025
Merged

Added possibility to generate base text on GPU for text evaluation. #1945

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1639e9d
Added possibility to generate on GPU.
andreyanufr Mar 19, 2025
546555f
1) Fixed merge conflict.
andreyanufr Mar 20, 2025
17537d7
Merge remote-tracking branch 'upstream/master' into aanuf/fix_gpu_gen…
andreyanufr Mar 21, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 13 additions & 18 deletions tools/who_what_benchmark/whowhatbench/text_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -189,29 +189,24 @@ def worst_examples(self, top_k: int = 5, metric="similarity"):
def _generate_data(self, model, gen_answer_fn=None, generation_config=None):
def default_gen_answer(model, tokenizer, prompt, max_new_tokens, crop_question, use_chat_template=False):
is_awq = getattr(model, "is_awq", None) is not None
device = "cpu"
if hasattr(model, "device"):
device = model.device

if use_chat_template:
message = [{"role": "user", "content": prompt}]
inputs = tokenizer.apply_chat_template(message, tokenize=True, add_generation_prompt=True, return_tensors="pt")
if is_awq:
with patch_awq_for_inference(is_awq):
tokens = model.generate(inputs, do_sample=False, max_new_tokens=max_new_tokens)
else:
tokens = model.generate(inputs, do_sample=False, max_new_tokens=max_new_tokens)
if crop_question:
tokens = tokens[:, inputs.shape[-1]:]
res = self.tokenizer.decode(tokens[0], skip_special_tokens=True)
return res
inputs = tokenizer.apply_chat_template(message, tokenize=True, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(device)
else:
inputs = self.tokenizer(prompt, return_tensors="pt")
if is_awq:
with patch_awq_for_inference(is_awq):
tokens = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens)
else:
inputs = self.tokenizer(prompt, return_tensors="pt").to(device)

if is_awq:
with patch_awq_for_inference(is_awq):
tokens = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens)
if crop_question:
tokens = tokens[:, inputs["input_ids"].shape[-1] :]
return self.tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]
else:
tokens = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens)
if crop_question:
tokens = tokens[:, inputs["input_ids"].shape[-1] :]
return self.tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]

gen_answer_fn = gen_answer_fn or default_gen_answer

Expand Down
Loading