huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-12 04:44:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Force ignore all images but last.

Browse Source
This commit is contained in:
Nicolas Patry 2024-10-01 10:18:41 +02:00
parent 265715a4f7
commit 7ede61bca6
No known key found for this signature in database
GPG Key ID: 64AF4752B2967863
1 changed files with 10 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
server/text_generation_server/models/mllama_causal_lm.py
View File

@ -91,9 +91,12 @@ class MllamaCausalLMBatch(VlmCausalLMBatch):
texts = [] texts = []
image_indices = [] image_indices = []
batch_tokenized_inputs = [] batch_tokenized_inputs = []
for i, r in enumerate(requests): for i, r in enumerate(requests):
# Each input is encoded into a list, where each element of this input list is either a string or a URL # Each input is encoded into a list, where each element of this input list is either a string or a URL
curr_text = "" curr_text = ""
curr_image = None
curr_i = None
for chunk in r.input_chunks.chunks: for chunk in r.input_chunks.chunks:
chunk_type = chunk.WhichOneof("chunk") chunk_type = chunk.WhichOneof("chunk")
if chunk_type == "text": if chunk_type == "text":
@ -103,11 +106,16 @@ class MllamaCausalLMBatch(VlmCausalLMBatch):
# TODO unsure about BOS # TODO unsure about BOS
curr_text += "<|image|>" curr_text += "<|image|>"
image_input = processor.image_processor(image, return_tensors="pt") image_input = processor.image_processor(image, return_tensors="pt")
image_inputs.append(image_input) curr_image = image_input
image_indices.append(i) curr_i = i
# image_inputs.append(image_input)
# image_indices.append(i)
else: else:
raise RuntimeError(f"Invalid chunk type {chunk_type}") raise RuntimeError(f"Invalid chunk type {chunk_type}")
texts.append(curr_text) texts.append(curr_text)
if curr_image is not None:
image_inputs.append(curr_image)
image_indices.append(curr_i)
input_ids = tokenizer( input_ids = tokenizer(
curr_text, curr_text,