Instructions to use irafm-llm/Recurrent-Llama-3.2-1B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use irafm-llm/Recurrent-Llama-3.2-1B with Transformers:

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

pipe = pipeline("text-generation", model="irafm-llm/Recurrent-Llama-3.2-1B", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("irafm-llm/Recurrent-Llama-3.2-1B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use irafm-llm/Recurrent-Llama-3.2-1B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "irafm-llm/Recurrent-Llama-3.2-1B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "irafm-llm/Recurrent-Llama-3.2-1B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/irafm-llm/Recurrent-Llama-3.2-1B

SGLang

How to use irafm-llm/Recurrent-Llama-3.2-1B 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 "irafm-llm/Recurrent-Llama-3.2-1B" \
    --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": "irafm-llm/Recurrent-Llama-3.2-1B",
		"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 "irafm-llm/Recurrent-Llama-3.2-1B" \
        --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": "irafm-llm/Recurrent-Llama-3.2-1B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use irafm-llm/Recurrent-Llama-3.2-1B with Docker Model Runner:
```
docker model run hf.co/irafm-llm/Recurrent-Llama-3.2-1B
```

Recurrent-Llama-3.2-1B / config.json

davidadamczyk

Recurrent-Llama-3.2-1B: surgery from Llama-3.2-1B + healing (FineWeb-Edu, ~98M tok)

eefd08c verified 3 months ago

Raw History Blame Contribute Delete

1.92 kB

	{
	"activation_checkpoint_impl": "per-iteration",
	"architecture_class_name": "RecurrentGPT",
	"architectures": [
	"RavenForCausalLM"
	],
	"attn_impl": "sdpa",
	"attn_logit_softcapping": null,
	"auto_map": {
	"AutoConfig": "raven_config_minimal.RavenConfig",
	"AutoModelForCausalLM": "raven_modeling_minimal.RavenForCausalLM"
	},
	"bias": false,
	"block_class_name": "SandwichBlock",
	"block_size": 4096,
	"block_type": "prenorm",
	"compare_mode": false,
	"effective_expected_depth": 200,
	"final_logit_softcapping": null,
	"head_dim": 64,
	"init_orthogonal": false,
	"init_strategy": "takase",
	"init_values": {
	"embed_scale": 1.0,
	"embedding": 0.013975424859373685,
	"out_proj": 0.0006987712429686843,
	"std": 0.013975424859373685
	},
	"injection_type": "linear",
	"intermediate_size": 8192,
	"max_position_embeddings": 131072,
	"mean_backprop_depth": 8,
	"mean_recurrence": 16,
	"mlp_class_name": "GatedMLP",
	"model_type": "huginn_raven",
	"n_embd": 2048,
	"n_heads": 32,
	"n_layers": 14,
	"n_layers_in_coda": 4,
	"n_layers_in_prelude": 4,
	"n_layers_in_recurrent_block": 6,
	"nonlin_name": "SiLU",
	"norm_class_name": "RMSNorm_llama",
	"norm_eps": 1e-05,
	"norm_type": "llama",
	"num_key_value_heads": 8,
	"padded_vocab_size": 128256,
	"padding_multiple": 4096,
	"qk_bias": false,
	"query_pre_attn_scalar": null,
	"rope_base": 500000.0,
	"rope_scaling": {
	"factor": 32.0,
	"high_freq_factor": 4.0,
	"low_freq_factor": 1.0,
	"original_max_position_embeddings": 8192,
	"rope_type": "llama3"
	},
	"rope_theta": 500000.0,
	"sampling_scheme": "poisson-lognormal-filling",
	"source_arch": "llama",
	"state_init": "like-init",
	"test_time_noise": 0,
	"test_time_noise_type": "fixed",
	"tie_embeddings": false,
	"tie_word_embeddings": false,
	"torch_dtype": "bfloat16",
	"transformers_version": "4.51.0",
	"vocab_size": 128256
	}