[Doc] Reorder vision language examples in alphabet order (#11228)
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py
@@ -33,78 +33,23 @@ class ModelRequestData(NamedTuple):
|
||||
# Unless specified, these settings have been tested to work on a single L4.
|
||||
|
||||
|
||||
def load_qwenvl_chat(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "Qwen/Qwen-VL-Chat"
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
trust_remote_code=True,
|
||||
max_model_len=1024,
|
||||
max_num_seqs=2,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
)
|
||||
placeholders = "".join(f"Picture {i}: <img></img>\n"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
|
||||
# This model does not have a chat_template attribute on its tokenizer,
|
||||
# so we need to explicitly pass it. We use ChatML since it's used in the
|
||||
# generation utils of the model:
|
||||
# https://huggingface.co/Qwen/Qwen-VL-Chat/blob/main/qwen_generation_utils.py#L265
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_name,
|
||||
trust_remote_code=True)
|
||||
|
||||
# Copied from: https://huggingface.co/docs/transformers/main/en/chat_templating
|
||||
chat_template = "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}" # noqa: E501
|
||||
|
||||
messages = [{'role': 'user', 'content': f"{placeholders}\n{question}"}]
|
||||
prompt = tokenizer.apply_chat_template(messages,
|
||||
tokenize=False,
|
||||
add_generation_prompt=True,
|
||||
chat_template=chat_template)
|
||||
|
||||
stop_tokens = ["<|endoftext|>", "<|im_start|>", "<|im_end|>"]
|
||||
stop_token_ids = [tokenizer.convert_tokens_to_ids(i) for i in stop_tokens]
|
||||
def load_aria(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "rhymes-ai/Aria"
|
||||
llm = LLM(model=model_name,
|
||||
tokenizer_mode="slow",
|
||||
trust_remote_code=True,
|
||||
dtype="bfloat16",
|
||||
limit_mm_per_prompt={"image": len(image_urls)})
|
||||
placeholders = "<fim_prefix><|img|><fim_suffix>\n" * len(image_urls)
|
||||
prompt = (f"<|im_start|>user\n{placeholders}{question}<|im_end|>\n"
|
||||
"<|im_start|>assistant\n")
|
||||
stop_token_ids = [93532, 93653, 944, 93421, 1019, 93653, 93519]
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=stop_token_ids,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=chat_template,
|
||||
)
|
||||
|
||||
|
||||
def load_phi3v(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
# num_crops is an override kwarg to the multimodal image processor;
|
||||
# For some models, e.g., Phi-3.5-vision-instruct, it is recommended
|
||||
# to use 16 for single frame scenarios, and 4 for multi-frame.
|
||||
#
|
||||
# Generally speaking, a larger value for num_crops results in more
|
||||
# tokens per image instance, because it may scale the image more in
|
||||
# the image preprocessing. Some references in the model docs and the
|
||||
# formula for image tokens after the preprocessing
|
||||
# transform can be found below.
|
||||
#
|
||||
# https://huggingface.co/microsoft/Phi-3.5-vision-instruct#loading-the-model-locally
|
||||
# https://huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/main/processing_phi3_v.py#L194
|
||||
llm = LLM(
|
||||
model="microsoft/Phi-3.5-vision-instruct",
|
||||
trust_remote_code=True,
|
||||
max_model_len=4096,
|
||||
max_num_seqs=2,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
mm_processor_kwargs={"num_crops": 4},
|
||||
)
|
||||
placeholders = "\n".join(f"<|image_{i}|>"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
prompt = f"<|user|>\n{placeholders}\n{question}<|end|>\n<|assistant|>\n"
|
||||
stop_token_ids = None
|
||||
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=stop_token_ids,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
chat_template=None)
|
||||
|
||||
|
||||
def load_h2onvl(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
@@ -141,6 +86,37 @@ def load_h2onvl(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
)
|
||||
|
||||
|
||||
def load_idefics3(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "HuggingFaceM4/Idefics3-8B-Llama3"
|
||||
|
||||
# The configuration below has been confirmed to launch on a single L40 GPU.
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
max_model_len=8192,
|
||||
max_num_seqs=16,
|
||||
enforce_eager=True,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
# if you are running out of memory, you can reduce the "longest_edge".
|
||||
# see: https://huggingface.co/HuggingFaceM4/Idefics3-8B-Llama3#model-optimizations
|
||||
mm_processor_kwargs={
|
||||
"size": {
|
||||
"longest_edge": 2 * 364
|
||||
},
|
||||
},
|
||||
)
|
||||
|
||||
placeholders = "\n".join(f"Image-{i}: <image>\n"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
prompt = f"<|begin_of_text|>User:{placeholders}\n{question}<end_of_utterance>\nAssistant:" # noqa: E501
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=None,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
|
||||
|
||||
def load_internvl(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "OpenGVLab/InternVL2-2B"
|
||||
|
||||
@@ -178,6 +154,28 @@ def load_internvl(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
)
|
||||
|
||||
|
||||
def load_mllama(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "meta-llama/Llama-3.2-11B-Vision-Instruct"
|
||||
|
||||
# The configuration below has been confirmed to launch on a single L40 GPU.
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
max_model_len=4096,
|
||||
max_num_seqs=16,
|
||||
enforce_eager=True,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
)
|
||||
|
||||
prompt = f"<|image|><|image|><|begin_of_text|>{question}"
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=None,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
|
||||
|
||||
def load_nvlm_d(question: str, image_urls: List[str]):
|
||||
model_name = "nvidia/NVLM-D-72B"
|
||||
|
||||
@@ -211,6 +209,80 @@ def load_nvlm_d(question: str, image_urls: List[str]):
|
||||
)
|
||||
|
||||
|
||||
def load_phi3v(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
# num_crops is an override kwarg to the multimodal image processor;
|
||||
# For some models, e.g., Phi-3.5-vision-instruct, it is recommended
|
||||
# to use 16 for single frame scenarios, and 4 for multi-frame.
|
||||
#
|
||||
# Generally speaking, a larger value for num_crops results in more
|
||||
# tokens per image instance, because it may scale the image more in
|
||||
# the image preprocessing. Some references in the model docs and the
|
||||
# formula for image tokens after the preprocessing
|
||||
# transform can be found below.
|
||||
#
|
||||
# https://huggingface.co/microsoft/Phi-3.5-vision-instruct#loading-the-model-locally
|
||||
# https://huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/main/processing_phi3_v.py#L194
|
||||
llm = LLM(
|
||||
model="microsoft/Phi-3.5-vision-instruct",
|
||||
trust_remote_code=True,
|
||||
max_model_len=4096,
|
||||
max_num_seqs=2,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
mm_processor_kwargs={"num_crops": 4},
|
||||
)
|
||||
placeholders = "\n".join(f"<|image_{i}|>"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
prompt = f"<|user|>\n{placeholders}\n{question}<|end|>\n<|assistant|>\n"
|
||||
stop_token_ids = None
|
||||
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=stop_token_ids,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
|
||||
|
||||
def load_qwenvl_chat(question: str, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "Qwen/Qwen-VL-Chat"
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
trust_remote_code=True,
|
||||
max_model_len=1024,
|
||||
max_num_seqs=2,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
)
|
||||
placeholders = "".join(f"Picture {i}: <img></img>\n"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
|
||||
# This model does not have a chat_template attribute on its tokenizer,
|
||||
# so we need to explicitly pass it. We use ChatML since it's used in the
|
||||
# generation utils of the model:
|
||||
# https://huggingface.co/Qwen/Qwen-VL-Chat/blob/main/qwen_generation_utils.py#L265
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_name,
|
||||
trust_remote_code=True)
|
||||
|
||||
# Copied from: https://huggingface.co/docs/transformers/main/en/chat_templating
|
||||
chat_template = "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}" # noqa: E501
|
||||
|
||||
messages = [{'role': 'user', 'content': f"{placeholders}\n{question}"}]
|
||||
prompt = tokenizer.apply_chat_template(messages,
|
||||
tokenize=False,
|
||||
add_generation_prompt=True,
|
||||
chat_template=chat_template)
|
||||
|
||||
stop_tokens = ["<|endoftext|>", "<|im_start|>", "<|im_end|>"]
|
||||
stop_token_ids = [tokenizer.convert_tokens_to_ids(i) for i in stop_tokens]
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=stop_token_ids,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=chat_template,
|
||||
)
|
||||
|
||||
|
||||
def load_qwen2_vl(question, image_urls: List[str]) -> ModelRequestData:
|
||||
try:
|
||||
from qwen_vl_utils import process_vision_info
|
||||
@@ -268,88 +340,16 @@ def load_qwen2_vl(question, image_urls: List[str]) -> ModelRequestData:
|
||||
)
|
||||
|
||||
|
||||
def load_mllama(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "meta-llama/Llama-3.2-11B-Vision-Instruct"
|
||||
|
||||
# The configuration below has been confirmed to launch on a single L40 GPU.
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
max_model_len=4096,
|
||||
max_num_seqs=16,
|
||||
enforce_eager=True,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
)
|
||||
|
||||
prompt = f"<|image|><|image|><|begin_of_text|>{question}"
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=None,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
|
||||
|
||||
def load_idefics3(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "HuggingFaceM4/Idefics3-8B-Llama3"
|
||||
|
||||
# The configuration below has been confirmed to launch on a single L40 GPU.
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
max_model_len=8192,
|
||||
max_num_seqs=16,
|
||||
enforce_eager=True,
|
||||
limit_mm_per_prompt={"image": len(image_urls)},
|
||||
# if you are running out of memory, you can reduce the "longest_edge".
|
||||
# see: https://huggingface.co/HuggingFaceM4/Idefics3-8B-Llama3#model-optimizations
|
||||
mm_processor_kwargs={
|
||||
"size": {
|
||||
"longest_edge": 2 * 364
|
||||
},
|
||||
},
|
||||
)
|
||||
|
||||
placeholders = "\n".join(f"Image-{i}: <image>\n"
|
||||
for i, _ in enumerate(image_urls, start=1))
|
||||
prompt = f"<|begin_of_text|>User:{placeholders}\n{question}<end_of_utterance>\nAssistant:" # noqa: E501
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=None,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None,
|
||||
)
|
||||
|
||||
|
||||
def load_aria(question, image_urls: List[str]) -> ModelRequestData:
|
||||
model_name = "rhymes-ai/Aria"
|
||||
llm = LLM(model=model_name,
|
||||
tokenizer_mode="slow",
|
||||
trust_remote_code=True,
|
||||
dtype="bfloat16",
|
||||
limit_mm_per_prompt={"image": len(image_urls)})
|
||||
placeholders = "<fim_prefix><|img|><fim_suffix>\n" * len(image_urls)
|
||||
prompt = (f"<|im_start|>user\n{placeholders}{question}<|im_end|>\n"
|
||||
"<|im_start|>assistant\n")
|
||||
stop_token_ids = [93532, 93653, 944, 93421, 1019, 93653, 93519]
|
||||
return ModelRequestData(
|
||||
llm=llm,
|
||||
prompt=prompt,
|
||||
stop_token_ids=stop_token_ids,
|
||||
image_data=[fetch_image(url) for url in image_urls],
|
||||
chat_template=None)
|
||||
|
||||
|
||||
model_example_map = {
|
||||
"phi3_v": load_phi3v,
|
||||
"h2ovl_chat": load_h2onvl,
|
||||
"internvl_chat": load_internvl,
|
||||
"NVLM_D": load_nvlm_d,
|
||||
"qwen2_vl": load_qwen2_vl,
|
||||
"qwen_vl_chat": load_qwenvl_chat,
|
||||
"mllama": load_mllama,
|
||||
"idefics3": load_idefics3,
|
||||
"aria": load_aria,
|
||||
"h2ovl_chat": load_h2onvl,
|
||||
"idefics3": load_idefics3,
|
||||
"internvl_chat": load_internvl,
|
||||
"mllama": load_mllama,
|
||||
"NVLM_D": load_nvlm_d,
|
||||
"phi3_v": load_phi3v,
|
||||
"qwen_vl_chat": load_qwenvl_chat,
|
||||
"qwen2_vl": load_qwen2_vl,
|
||||
}
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user