Instructions to use allura-forge/claumba-micro-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use allura-forge/claumba-micro-sft with Transformers:

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

pipe = pipeline("text-generation", model="allura-forge/claumba-micro-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("allura-forge/claumba-micro-sft")
model = AutoModelForCausalLM.from_pretrained("allura-forge/claumba-micro-sft")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use allura-forge/claumba-micro-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "allura-forge/claumba-micro-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "allura-forge/claumba-micro-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/allura-forge/claumba-micro-sft

SGLang

How to use allura-forge/claumba-micro-sft 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 "allura-forge/claumba-micro-sft" \
    --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": "allura-forge/claumba-micro-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "allura-forge/claumba-micro-sft" \
        --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": "allura-forge/claumba-micro-sft",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use allura-forge/claumba-micro-sft with Docker Model Runner:
```
docker model run hf.co/allura-forge/claumba-micro-sft
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

See axolotl config

axolotl version: 0.13.0.dev0

# ------------------------------------------------------------------
# 0.  Model & Tokeniser
# ------------------------------------------------------------------
base_model:     ibm-granite/granite-4.0-h-micro
trust_remote_code: true

# ------------------------------------------------------------------
# 1.  Precision & Memory
# ------------------------------------------------------------------
bf16:  auto
fp16:
tf32:  false

load_in_8bit:  false
load_in_4bit:  false

# vram helpers
flash_attention:  true
# gradient_checkpointing: true   # <-- uncomment if you want old-style GC instead of FSDP AC

# ------------------------------------------------------------------
# 2.  FSDP (zero-3 + cpu-offload)
# ------------------------------------------------------------------
fsdp:
  - auto_wrap
  - full_shard

fsdp_config:
  fsdp_version: 2
  fsdp_offload_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: GraniteMoeHybridDecoderLayer
  fsdp_state_dict_type: SHARDED_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD
  fsdp_reshard_after_forward: true
  fsdp_activation_checkpointing: true     # disables itself if unsupported

# ------------------------------------------------------------------
# 3.  Training Schedule
# ------------------------------------------------------------------
num_epochs: 2
learning_rate: 2e-5
lr_scheduler: cosine
warmup_ratio: 0.05
max_grad_norm: 0.1
weight_decay: 0.0
optimizer: adamw_torch_8bit

micro_batch_size: 2
gradient_accumulation_steps: 2
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

# saves / eval frequency
saves_per_epoch: 4
val_set_size: 0.0
logging_steps: 1
strict: false

# ------------------------------------------------------------------
# 4.  Data & Prompt Template
# ------------------------------------------------------------------
datasets:
  - path: allura-forge/claude-oss-sft
    type: chat_template
    split: train
    field_messages: conversations
    message_field_role: from
    message_field_content: value

chat_template: jinja
chat_template_jinja: |
  {%- for message in messages -%}
    {{- '<|start_of_role|>' + message['role'] + '<|end_of_role|>' + message['content'] + '<|end_of_text|>' -}}
    {%- if loop.last and add_generation_prompt -%}
        {{- '<|start_of_role|>assistant<|end_of_role|>' -}}
    {%- endif -%}
  {%- endfor -%}

shuffle_merged_datasets: true
dataset_prepared_path: last_run_prepared
remove_unused_columns: false
train_on_inputs: false
group_by_length: false

# ------------------------------------------------------------------
# 5.  Plug-ins (memory / speed)
# ------------------------------------------------------------------
plugins:
  - axolotl.integrations.liger.LigerPlugin

# ------------------------------------------------------------------
# 6.  Weights & Biases
# ------------------------------------------------------------------
wandb_project: claumba-micro
wandb_name: woke
wandb_entity:
wandb_watch:
wandb_log_model:

# ------------------------------------------------------------------
# 7.  I/O & Resume
# ------------------------------------------------------------------
output_dir: ./model-output
resume_from_checkpoint:
local_rank:

# ------------------------------------------------------------------
# 8.  Unused / commented-out
# ------------------------------------------------------------------
# evals_per_epoch:
# eval_steps: 100
# eval_sample_packing: false
# early_stopping_patience:
# xformers_attention:

model-output

This model is a fine-tuned version of ibm-granite/granite-4.0-h-micro on the allura-forge/claude-oss-sft dataset.

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: 2e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 2
total_train_batch_size: 32
total_eval_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_8BIT 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: 15
training_steps: 308

Training results

Framework versions

Transformers 4.57.1
Pytorch 2.8.0+cu128
Datasets 4.0.0
Tokenizers 0.22.1

Downloads last month: 27

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for allura-forge/claumba-micro-sft

Base model

ibm-granite/granite-4.0-h-micro

Finetuned

(10)

this model

Quantizations

1 model

allura-forge
/

claumba-micro-sft

model-output

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Model tree for allura-forge/claumba-micro-sft

Dataset used to train allura-forge/claumba-micro-sft