Skip to content

llm_compressor example of llama 4 scout #80

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

Merged
merged 1 commit into from
Nov 5, 2025
Merged

llm_compressor example of llama 4 scout #80

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
15 changes: 11 additions & 4 deletions hf_torchao_vllm/quantize_hf_model_with_llm_compressor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,6 @@
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.utils import dispatch_for_generation

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer


Expand All @@ -17,9 +16,7 @@ def run(
assert quant_type in ("fp8", "nvfp4"), "unsupported"

# Load model.
model = AutoModelForCausalLM.from_pretrained(
model_name, torch_dtype=torch.bfloat16
)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto")
print(model)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Expand All @@ -34,6 +31,16 @@ def run(
"re:.*shared_expert.*",
]
)
elif model_name == "meta-llama/Llama-4-Scout-17B-16E-Instruct":
ignore_list.extend(
[
"re:.*self_attn",
"re:.*router",
"re:.*vision_model.*",
"re:.*multi_modal_projector.*",
"Llama4TextAttention",
]
)

if quant_type == "fp8":
# Configure the quantization algorithm and scheme.
Expand Down