Image-Text-to-Text
Transformers
TensorBoard
Safetensors
gemma4
Generated from Trainer
conversational
Instructions to use AlexHung29629/MyEBFTModelSFTCP with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use AlexHung29629/MyEBFTModelSFTCP with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="AlexHung29629/MyEBFTModelSFTCP") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("AlexHung29629/MyEBFTModelSFTCP") model = AutoModelForMultimodalLM.from_pretrained("AlexHung29629/MyEBFTModelSFTCP", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use AlexHung29629/MyEBFTModelSFTCP with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "AlexHung29629/MyEBFTModelSFTCP" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "AlexHung29629/MyEBFTModelSFTCP", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/AlexHung29629/MyEBFTModelSFTCP
- SGLang
How to use AlexHung29629/MyEBFTModelSFTCP with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "AlexHung29629/MyEBFTModelSFTCP" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "AlexHung29629/MyEBFTModelSFTCP", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "AlexHung29629/MyEBFTModelSFTCP" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "AlexHung29629/MyEBFTModelSFTCP", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use AlexHung29629/MyEBFTModelSFTCP with Docker Model Runner:
docker model run hf.co/AlexHung29629/MyEBFTModelSFTCP
See axolotl config
axolotl version: 0.21.1
# 4 節點 × 8 張 H100,全參數 FSDP2。序列 131072,Ulysses CP 2。
# 資料是 train-clean-noeaft-cp.yml 的三份 clean jsonl,加上 train-search-noeaft-cp.yml 的 search。
# 超參、CP、CCE、FSDP 跟那兩份相同。不要改那兩份,也不要覆寫它們的 output_dir。
# 權重起點是 AlexHung29629/ebft_gemma4_26b_lora_lr5e6_merged。
# snapshot 已在 HF_HOME。worker 設 HF_HUB_OFFLINE=1,不要改去 google/gemma-4-26B-A4B-it。
# 這是新的 run:新的 output_dir、新的 prepared cache、新的 cosine。不能從舊 optimizer 續訓。
# 由 prep-mix-ebft-noeaft-cp.sbatch 先 preprocess,再由 train-mix-ebft-noeaft-cp.sbatch 送出。
# 容器是 ../axolotl-cp.sqsh。
# 不要開 use_eaft。183855 已確認 EAFT 不能跟 context parallel 同時開。
# 不要在登入節點跑 preprocess 或 train。
# 同一個 CP group 的 2 張卡看同一條序列。
# global batch = (32 / context_parallel.size) * micro_batch_size * gradient_accumulation_steps。
base_model: AlexHung29629/ebft_gemma4_26b_lora_lr5e6_merged
# Gemma 4 權重是 Gemma4ForConditionalGeneration,不要強制 AutoModelForCausalLM。
# 這次是文字 SFT,凍結 vision / audio tower。
freeze_mm_modules: true
plugins:
- axolotl.integrations.liger.LigerPlugin
- axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
# multimodal Gemma 4 的 liger 沒有 fused linear CE。
liger_fused_linear_cross_entropy: false
# 純交叉熵。CCE 不物化整段 logits,並套用 final_logit_softcapping=30。
# CP 2 時每張卡看 65536 token。這段 bf16 logits 約 32GiB,放不進 H100。
# 不要開 use_eaft。183855 已確認 EAFT 不能跟 context parallel 同時開。
cut_cross_entropy: true
use_eaft: false
# 全參數。不要加 adapter、4bit、quantize_moe_experts。
load_in_8bit: false
load_in_4bit: false
datasets:
- path: data/ultradata-sft-clean.jsonl
type: input_output
- path: data/ultradata-sft-agent-clean.jsonl
type: input_output
- path: data/tw-sft-clean.jsonl
type: input_output
- path: data/search-sft-clean.jsonl
type: input_output
dataset_prepared_path: ./prepared/gemma4-26b-ebft-mix-131072
output_dir: ./outputs/gemma4-26b-ebft-mix-noeaft-cp
dataset_num_proc: 64
train_on_inputs: false
val_set_size: 0
seed: 42
# 128k。超過這個長度的樣本,Axolotl preprocess 會丟掉。
sequence_len: 131072
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
use_tensorboard: true
# 32 張 H100,CP 2。每個 step 16 條序列。
# global batch = 32 / 2 * 1 * 1 * 131072 = 2,097,152 tokens。
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_torch_fused
learning_rate: 5.0e-6
lr_scheduler: cosine
warmup_ratio: 0.03
weight_decay: 0.01
max_grad_norm: 1.0
bf16: auto
tf32: true
gradient_checkpointing: true
gradient_checkpointing_kwargs:
use_reentrant: false
# 183274 打開後,backward 仍是已佔 72.23GiB 再要 5.50GiB。這段 workspace 沒被 offload 移走。
activation_offloading: false
# FSDP2 把參數、梯度、optimizer state 切到 32 張卡(dp_shard × cp)。
# 必須包文字 decoder。不指定時 FSDP 會去找不存在的 Gemma4AudioLayer。
# 存檔用 FULL_STATE_DICT,集中成一份權重。不寫 optimizer state,不能從 optimizer 續訓。
# 185902:Accelerate 只從環境變數組 mesh。context_parallel 會設 CP_SIZE=2,
# dp_shard 預設 1,total_size=2,對不上 32 個 process。
# 明寫 dp_shard_size,axolotl 才會設 PARALLELISM_CONFIG_DP_SHARD_SIZE。
# 32 卡 / CP 2 = 16。改節點數或 CP 時要一起改。
dp_shard_size: 16
fsdp_version: 2
fsdp_config:
offload_params: false
cpu_ram_efficient_loading: true
auto_wrap_policy: TRANSFORMER_BASED_WRAP
transformer_layer_cls_to_wrap: Gemma4TextDecoderLayer
state_dict_type: FULL_STATE_DICT
reshard_after_forward: true
# Ulysses 把序列切成 2 段。auto 搭 flex 會被擋,backend 一定要明寫 ulysses。
# 不要加 gemma4_hybrid_attn_impl,它會繞過 CP wrapper。
# 185909 的 mesh 已是 (16, 2)。接著 CP 讀 num_key_value_heads,
# Gemma 4 sliding=8、full attention=2,Transformers 拒絕讀全域值。
# 185910:Ulysses create_block_mask 會實體化整段 Q×KV,131072 的 int64 是 128GiB。
# 這兩件事都由 cp_compat/sitecustomize.py 處理,train_worker 用 PYTHONPATH 掛上。
context_parallel:
size: 2
backend: ulysses
# flex attention。全域層 head_dim=512,FA2 單獨跑不了。
# 參數照 MEGA/train.yml。H100 預設 flex 會 SRAM OOM。Ulysses 的 flex 也吃這組。
attn_implementation: flex_attention
flex_attn_kernel_options:
num_stages: 1
bwd_num_stages: 1
logging_steps: 1
saves_per_epoch: 1
save_total_limit: 2
# 只存模型參數。optimizer、scheduler、rng 都不寫進 checkpoint。
save_only_model: true
special_tokens:
bos_token: "<bos>"
eos_token: "<eos>"
pad_token: "<pad>"
unk_token: "<unk>"
# preprocess --debug 若把要學的 <turn|> 標成 -100,拿掉這段。
eot_tokens:
- "<turn|>"
outputs/gemma4-26b-ebft-mix-noeaft-cp
This model is a fine-tuned version of AlexHung29629/ebft_gemma4_26b_lora_lr5e6_merged on the data/ultradata-sft-clean.jsonl, the data/ultradata-sft-agent-clean.jsonl, the data/tw-sft-clean.jsonl and the data/search-sft-clean.jsonl datasets.
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 32
- total_train_batch_size: 32
- total_eval_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 33
- training_steps: 1122
Training results
Framework versions
- Transformers 5.20.1
- Pytorch 2.13.0+cu130
- Datasets 5.0.1
- Tokenizers 0.23.3
- Downloads last month
- 1
Model tree for AlexHung29629/MyEBFTModelSFTCP
Base model
google/gemma-4-26B-A4B