Spaces:
Sleeping
Sleeping
Tuchuanhuhuhu
commited on
Commit
·
53518f7
1
Parent(s):
67d913f
优化llama模型
Browse files- modules/models.py +17 -9
- requirements_advanced.txt +1 -1
modules/models.py
CHANGED
@@ -340,9 +340,14 @@ class LLaMA_Client(BaseLLMModel):
|
|
340 |
self.end_string = "\n\n"
|
341 |
|
342 |
def _get_llama_style_input(self):
|
343 |
-
history = [
|
344 |
-
|
345 |
-
|
|
|
|
|
|
|
|
|
|
|
346 |
return context
|
347 |
|
348 |
def get_answer_at_once(self):
|
@@ -365,14 +370,15 @@ class LLaMA_Client(BaseLLMModel):
|
|
365 |
def get_answer_stream_iter(self):
|
366 |
context = self._get_llama_style_input()
|
367 |
partial_text = ""
|
368 |
-
|
|
|
369 |
input_dataset = self.dataset.from_dict(
|
370 |
{"type": "text_only", "instances": [{"text": context+partial_text}]}
|
371 |
)
|
372 |
output_dataset = self.inferencer.inference(
|
373 |
model=self.model,
|
374 |
dataset=input_dataset,
|
375 |
-
max_new_tokens=
|
376 |
temperature=self.temperature,
|
377 |
)
|
378 |
response = output_dataset.to_dict()["instances"][0]["text"]
|
@@ -402,9 +408,11 @@ class ModelManager:
|
|
402 |
dont_change_lora_selector = False
|
403 |
if model_type != ModelType.OpenAI:
|
404 |
config.local_embedding = True
|
|
|
405 |
model = None
|
406 |
try:
|
407 |
if model_type == ModelType.OpenAI:
|
|
|
408 |
model = OpenAIClient(
|
409 |
model_name=model_name,
|
410 |
api_key=access_key,
|
@@ -413,15 +421,17 @@ class ModelManager:
|
|
413 |
top_p=top_p,
|
414 |
)
|
415 |
elif model_type == ModelType.ChatGLM:
|
|
|
416 |
model = ChatGLM_Client(model_name)
|
417 |
elif model_type == ModelType.LLaMA and lora_model_path == "":
|
418 |
-
msg = "
|
419 |
logging.info(msg)
|
420 |
lora_selector_visibility = True
|
421 |
if os.path.isdir("lora"):
|
422 |
lora_choices = get_file_names("lora", plain=True, filetypes=[""])
|
423 |
lora_choices = ["No LoRA"] + lora_choices
|
424 |
elif model_type == ModelType.LLaMA and lora_model_path != "":
|
|
|
425 |
dont_change_lora_selector = True
|
426 |
if lora_model_path == "No LoRA":
|
427 |
lora_model_path = None
|
@@ -429,15 +439,13 @@ class ModelManager:
|
|
429 |
else:
|
430 |
msg += f" + {lora_model_path}"
|
431 |
model = LLaMA_Client(model_name, lora_model_path)
|
432 |
-
pass
|
433 |
elif model_type == ModelType.Unknown:
|
434 |
raise ValueError(f"未知模型: {model_name}")
|
435 |
logging.info(msg)
|
436 |
except Exception as e:
|
437 |
logging.error(e)
|
438 |
msg = f"{STANDARD_ERROR_MSG}: {e}"
|
439 |
-
|
440 |
-
self.model = model
|
441 |
if dont_change_lora_selector:
|
442 |
return msg
|
443 |
else:
|
|
|
340 |
self.end_string = "\n\n"
|
341 |
|
342 |
def _get_llama_style_input(self):
|
343 |
+
history = []
|
344 |
+
for x in self.history:
|
345 |
+
if x["role"] == "user":
|
346 |
+
history.append(f"Input: {x['content']}")
|
347 |
+
else:
|
348 |
+
history.append(f"Output: {x['content']}")
|
349 |
+
context = "\n\n".join(history)
|
350 |
+
context += "\n\nOutput: "
|
351 |
return context
|
352 |
|
353 |
def get_answer_at_once(self):
|
|
|
370 |
def get_answer_stream_iter(self):
|
371 |
context = self._get_llama_style_input()
|
372 |
partial_text = ""
|
373 |
+
step = 1
|
374 |
+
for _ in range(0, self.max_generation_token, step):
|
375 |
input_dataset = self.dataset.from_dict(
|
376 |
{"type": "text_only", "instances": [{"text": context+partial_text}]}
|
377 |
)
|
378 |
output_dataset = self.inferencer.inference(
|
379 |
model=self.model,
|
380 |
dataset=input_dataset,
|
381 |
+
max_new_tokens=step,
|
382 |
temperature=self.temperature,
|
383 |
)
|
384 |
response = output_dataset.to_dict()["instances"][0]["text"]
|
|
|
408 |
dont_change_lora_selector = False
|
409 |
if model_type != ModelType.OpenAI:
|
410 |
config.local_embedding = True
|
411 |
+
self.model = None
|
412 |
model = None
|
413 |
try:
|
414 |
if model_type == ModelType.OpenAI:
|
415 |
+
logging.info(f"正在加载OpenAI模型: {model_name}")
|
416 |
model = OpenAIClient(
|
417 |
model_name=model_name,
|
418 |
api_key=access_key,
|
|
|
421 |
top_p=top_p,
|
422 |
)
|
423 |
elif model_type == ModelType.ChatGLM:
|
424 |
+
logging.info(f"正在加载ChatGLM模型: {model_name}")
|
425 |
model = ChatGLM_Client(model_name)
|
426 |
elif model_type == ModelType.LLaMA and lora_model_path == "":
|
427 |
+
msg = f"现在请为 {model_name} 选择LoRA模型"
|
428 |
logging.info(msg)
|
429 |
lora_selector_visibility = True
|
430 |
if os.path.isdir("lora"):
|
431 |
lora_choices = get_file_names("lora", plain=True, filetypes=[""])
|
432 |
lora_choices = ["No LoRA"] + lora_choices
|
433 |
elif model_type == ModelType.LLaMA and lora_model_path != "":
|
434 |
+
logging.info(f"正在加载LLaMA模型: {model_name} + {lora_model_path}")
|
435 |
dont_change_lora_selector = True
|
436 |
if lora_model_path == "No LoRA":
|
437 |
lora_model_path = None
|
|
|
439 |
else:
|
440 |
msg += f" + {lora_model_path}"
|
441 |
model = LLaMA_Client(model_name, lora_model_path)
|
|
|
442 |
elif model_type == ModelType.Unknown:
|
443 |
raise ValueError(f"未知模型: {model_name}")
|
444 |
logging.info(msg)
|
445 |
except Exception as e:
|
446 |
logging.error(e)
|
447 |
msg = f"{STANDARD_ERROR_MSG}: {e}"
|
448 |
+
self.model = model
|
|
|
449 |
if dont_change_lora_selector:
|
450 |
return msg
|
451 |
else:
|
requirements_advanced.txt
CHANGED
@@ -2,6 +2,6 @@ transformers
|
|
2 |
torch
|
3 |
icetk
|
4 |
protobuf==3.19.0
|
5 |
-
git+https://github.com/OptimalScale/LMFlow.git
|
6 |
cpm-kernels
|
7 |
sentence_transformers
|
|
|
2 |
torch
|
3 |
icetk
|
4 |
protobuf==3.19.0
|
5 |
+
git+https://github.com/OptimalScale/LMFlow.git
|
6 |
cpm-kernels
|
7 |
sentence_transformers
|