Instructions to use giux78/zagreus-test-202000-sft-2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use giux78/zagreus-test-202000-sft-2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="giux78/zagreus-test-202000-sft-2")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("giux78/zagreus-test-202000-sft-2")
model = AutoModelForCausalLM.from_pretrained("giux78/zagreus-test-202000-sft-2")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use giux78/zagreus-test-202000-sft-2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "giux78/zagreus-test-202000-sft-2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "giux78/zagreus-test-202000-sft-2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/giux78/zagreus-test-202000-sft-2

SGLang

How to use giux78/zagreus-test-202000-sft-2 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 "giux78/zagreus-test-202000-sft-2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "giux78/zagreus-test-202000-sft-2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "giux78/zagreus-test-202000-sft-2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "giux78/zagreus-test-202000-sft-2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use giux78/zagreus-test-202000-sft-2 with Docker Model Runner:
```
docker model run hf.co/giux78/zagreus-test-202000-sft-2
```

See axolotl config

axolotl version: 0.9.2

base_model: giux78/zagreus-test-202000
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

#tokenizer_type: auto
strict: false
seed: 1337
output_dir: ./ale_outputs/zagreus-350M-sft

# === Datasets ===
streaming: false
datasets:
  - path: /leonardo_work/EUHPC_A04_045/training/sft_data #/leonardo_work/EUHPC_A04_045/training/test_data #/leonardo_work/EUHPC_A04_045/.data
    type: chat_template
    chat_template: tokenizer_default_fallback_chatml
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
    roles:
      user: ["human","user"]
      assistant: ["gpt","assistant"]
      system: ["system"]
      tool: ["tool"]
    roles_to_train: ["assistant"]
    train_on_eos: "turn"

# === Sequencing / packing ===
sequence_len: 4096
sample_packing: true
remove_unused_columns: false   # <-- aggiungi questa riga
eval_sample_packing: false
pad_to_sequence_len: false
streaming_multipack_buffer_size: 10000

# === Ottimizzazione ===
#optimizer: adamw_torch_fused
#learning_rate: 2e-5
#lr_scheduler: cosine
#warmup_ratio: 0.1
#weight_decay: 0.0

optimizer: adamw_torch_fused
learning_rate: 5e-4              # ~scaling lineare per eff. batch 512
lr_scheduler: cosine
warmup_ratio: 0.01
weight_decay: 0.10
adam_beta1: 0.9
adam_beta2: 0.95
adam_epsilon: 1e-8
max_grad_norm: 1.0


# === Batch (per GPU) ===
micro_batch_size: 1
gradient_accumulation_steps: 16 #8
# Eff. batch = micro_batch_size * grad_accum * num_gpus = 1 * 8 * 32 = 256

# === Precisione / memoria ===
bf16: auto
tf32: true
flash_attention: true
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false

# === FSDP (Axolotl usa fsdp_config; la chiave "fsdp:" è deprecata) ===
fsdp_config:
  fsdp_sharding_strategy: FULL_SHARD           # shard di param, grad e optimizer state
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
  fsdp_use_orig_params: false
  fsdp_sync_module_states: true
  fsdp_limit_all_gathers: true
  fsdp_cpu_ram_efficient_loading: true
  fsdp_offload_params: false                   # attivalo solo se VRAM è stretta (vedi variante)
  fsdp_state_dict_type: SHARDED_STATE_DICT     # checkpoint più leggeri su cluster multi-nodo

# === Loop di training ===
num_epochs: 1                                  # con 170GB basta una passata
# max_steps: 200000                            # alternativa: budget a step/token

# === Eval / checkpoint ===
val_set_size: 0.01
evals_per_epoch: 5
save_steps: 1000          # salva ogni 2.000 step (metti il valore che preferisci)
save_total_limit: 5
logging_steps: 20


# === Tracciamento ===
wandb_mode: "offline"
wandb_project: zagreus-350M-sft
wandb_entity: mii-llm
wandb_name: sft

# === Token speciali ===
special_tokens:
  bos_token: <|begin_of_text|>
  pad_token: <|end_of_text|>
  eos_token: <|end_of_text|>

ale_outputs/zagreus-350M-sft

This model is a fine-tuned version of giux78/zagreus-test-202000 on the None dataset. It achieves the following results on the evaluation set:

Loss: nan

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: 0.0005
train_batch_size: 1
eval_batch_size: 1
seed: 1337
distributed_type: multi-GPU
num_devices: 32
gradient_accumulation_steps: 16
total_train_batch_size: 512
total_eval_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 2
num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0055	1	nan
3.1174	0.2028	37	nan
2.0899	0.4056	74	nan
1.9786	0.6084	111	nan
1.9433	0.8112	148	nan

Framework versions

Transformers 4.56.2
Pytorch 2.5.1+cu121
Datasets 3.5.1
Tokenizers 0.22.1

Downloads last month: 1

Safetensors

Model size

0.4B params

Tensor type

BF16

Model tree for giux78/zagreus-test-202000-sft-2

Base model

giux78/zagreus-test-202000

Finetuned

(12)

this model