docs: add canonical tooling corpus (147 files) from Google/HF/frameworks
Five-lane parallel research pass. Each subdir under tooling/ has its own README indexing downloaded files with verified upstream sources. - google-official/: deepmind-gemma JAX examples, gemma_pytorch scripts, gemma.cpp API server docs, google-gemma/cookbook notebooks, ai.google.dev HTML snapshots, Gemma 3 tech report - huggingface/: 8 gemma-4-* model cards, chat-template .jinja files, tokenizer_config.json, transformers gemma4/ source, launch blog posts, official HF Spaces app.py - inference-frameworks/: vLLM/llama.cpp/MLX/Keras-hub/TGI/Gemini API/Vertex AI comparison, run_commands.sh with 8 working launches, 9 code snippets - gemma-family/: 12 per-variant briefs (ShieldGemma 2, CodeGemma, PaliGemma 2, Recurrent/Data/Med/TxGemma, Embedding/Translate/Function/Dolphin/SignGemma) - fine-tuning/: Unsloth Gemma 4 notebooks, Axolotl YAMLs (incl 26B-A4B MoE), TRL scripts, Google cookbook fine-tune notebooks, recipe-recommendation.md Findings that update earlier CORPUS_* docs are flagged in tooling/README.md (not applied) — notably the new <|turn>/<turn|> prompt format, gemma_pytorch abandonment, gemma.cpp Gemini-API server, transformers AutoModelForMultimodalLM, FA2 head_dim=512 break, 26B-A4B MoE quantization rules, no Gemma 4 tech report PDF yet, no Gemma-4-generation specialized siblings yet. Pre-commit secrets hook bypassed per user authorization — flagged "secrets" are base64 notebook cell outputs and example Ed25519 keys in the HDP agentic-security demo, not real credentials. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Mortdecai
@@ -0,0 +1,157 @@
|
||||
# Copyright 2020-2026 The HuggingFace Team. All rights reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
|
||||
# /// script
|
||||
# dependencies = [
|
||||
# "trl[peft]",
|
||||
# "Pillow",
|
||||
# "math-verify",
|
||||
# "latex2sympy2_extended",
|
||||
# "torchvision",
|
||||
# "trackio",
|
||||
# "kernels",
|
||||
# ]
|
||||
# ///
|
||||
|
||||
"""
|
||||
pip install math_verify
|
||||
|
||||
# For Qwen/Qwen2.5-VL-3B-Instruct
|
||||
accelerate launch \
|
||||
--config_file examples/accelerate_configs/deepspeed_zero3.yaml \
|
||||
examples/scripts/grpo_vlm.py \
|
||||
--model_name_or_path Qwen/Qwen2.5-VL-3B-Instruct \
|
||||
--output_dir grpo-Qwen2.5-VL-3B-Instruct \
|
||||
--learning_rate 1e-5 \
|
||||
--dtype bfloat16 \
|
||||
--max_completion_length 1024 \
|
||||
--use_vllm \
|
||||
--vllm_mode colocate \
|
||||
--use_peft \
|
||||
--lora_target_modules "q_proj", "v_proj" \
|
||||
--log_completions
|
||||
|
||||
# For HuggingFaceTB/SmolVLM2-2.2B-Instruct
|
||||
pip install num2words==0.5.14
|
||||
|
||||
accelerate launch \
|
||||
--config_file examples/accelerate_configs/deepspeed_zero3.yaml \
|
||||
examples/scripts/grpo_vlm.py \
|
||||
--model_name_or_path HuggingFaceTB/SmolVLM2-2.2B-Instruct \
|
||||
--output_dir grpo-SmolVLM2-2.2B-Instruct \
|
||||
--learning_rate 1e-5 \
|
||||
--dtype bfloat16 \
|
||||
--max_completion_length 1024 \
|
||||
--use_peft \
|
||||
--lora_target_modules "q_proj", "v_proj" \
|
||||
--log_completions \
|
||||
--per_device_train_batch_size 1 \
|
||||
--gradient_accumulation_steps 2 \
|
||||
--num_generations 2
|
||||
|
||||
"""
|
||||
|
||||
import torch
|
||||
from datasets import load_dataset
|
||||
|
||||
from trl import (
|
||||
GRPOConfig,
|
||||
GRPOTrainer,
|
||||
ModelConfig,
|
||||
ScriptArguments,
|
||||
TrlParser,
|
||||
get_kbit_device_map,
|
||||
get_peft_config,
|
||||
get_quantization_config,
|
||||
)
|
||||
from trl.rewards import accuracy_reward, think_format_reward
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = TrlParser((ScriptArguments, GRPOConfig, ModelConfig))
|
||||
script_args, training_args, model_args = parser.parse_args_and_config()
|
||||
################
|
||||
# Model
|
||||
################
|
||||
dtype = model_args.dtype if model_args.dtype in ["auto", None] else getattr(torch, model_args.dtype)
|
||||
training_args.model_init_kwargs = dict(
|
||||
revision=model_args.model_revision,
|
||||
attn_implementation=model_args.attn_implementation,
|
||||
dtype=dtype,
|
||||
)
|
||||
quantization_config = get_quantization_config(model_args)
|
||||
if quantization_config is not None:
|
||||
# Passing None would not be treated the same as omitting the argument, so we include it only when valid.
|
||||
training_args.model_init_kwargs["device_map"] = get_kbit_device_map()
|
||||
training_args.model_init_kwargs["quantization_config"] = quantization_config
|
||||
|
||||
################
|
||||
# Dataset
|
||||
################
|
||||
dataset = load_dataset("lmms-lab/multimodal-open-r1-8k-verified", split="train")
|
||||
dataset = dataset.train_test_split(test_size=100, seed=42)
|
||||
|
||||
SYSTEM_PROMPT = (
|
||||
"A conversation between user and assistant. The user asks a question, and the assistant solves it. The "
|
||||
"assistant first thinks about the reasoning process in the mind and then provides the user with the answer. "
|
||||
"The reasoning process and answer are enclosed within <think></think> tags, i.e., <think>\nThis is my "
|
||||
"reasoning.\n</think>\nThis is my answer."
|
||||
)
|
||||
|
||||
def make_conversation(example):
|
||||
prompt = [
|
||||
{"role": "system", "content": SYSTEM_PROMPT},
|
||||
{"role": "user", "content": example["problem"]},
|
||||
]
|
||||
return {"prompt": prompt}
|
||||
|
||||
dataset = dataset.map(make_conversation)
|
||||
|
||||
# Filter have big images
|
||||
def filter_big_images(example):
|
||||
image = example["image"]
|
||||
return image.size[0] < 512 and image.size[1] < 512
|
||||
|
||||
dataset = dataset.filter(filter_big_images)
|
||||
|
||||
def convert_to_rgb(example):
|
||||
image = example["image"]
|
||||
if image.mode != "RGB":
|
||||
image = image.convert("RGB")
|
||||
example["image"] = image
|
||||
return example
|
||||
|
||||
dataset = dataset.map(convert_to_rgb)
|
||||
|
||||
train_dataset = dataset["train"]
|
||||
eval_dataset = dataset["test"] if training_args.eval_strategy != "no" else None
|
||||
|
||||
################
|
||||
# Training
|
||||
################
|
||||
trainer = GRPOTrainer(
|
||||
model=model_args.model_name_or_path,
|
||||
args=training_args,
|
||||
reward_funcs=[think_format_reward, accuracy_reward],
|
||||
train_dataset=train_dataset,
|
||||
eval_dataset=eval_dataset,
|
||||
peft_config=get_peft_config(model_args),
|
||||
)
|
||||
|
||||
trainer.train()
|
||||
|
||||
# Save and push to hub
|
||||
trainer.save_model(training_args.output_dir)
|
||||
if training_args.push_to_hub:
|
||||
trainer.push_to_hub(dataset_name=script_args.dataset_name)
|
||||
Reference in New Issue
Block a user