{
  "hf_id": "LiquidAI/LFM2.5-1.2B-Base",
  "meta": {
    "title": "LFM2.5 1.2B Base",
    "slug": "lfm2.5-1.2b-base",
    "provider": "LiquidAI",
    "description": "Liquid AI's 1.2B pretrained base model on the LFM2 hybrid conv+attention backbone — a text-completion and fine-tuning foundation (no chat template).",
    "date_added": "2026-06-25",
    "date_updated": "2026-06-24",
    "tasks": [
      "text"
    ],
    "performance_headline": "1.2B pretrained base (no chat template) — completions endpoint, ideal as a fine-tuning base",
    "related_recipes": [
      "LiquidAI/LFM2.5-1.2B-Instruct",
      "LiquidAI/LFM2.5-350M"
    ],
    "hardware": {
      "h100": "verified",
      "h200": "verified",
      "b200": "verified"
    }
  },
  "recommended_command": {
    "hardware": "h200",
    "strategy": "single_node_tp",
    "variant": "default",
    "node_count": 1,
    "deploy_type": "single_node",
    "env": {},
    "docker_image": "vllm/vllm-openai:latest",
    "command": "vllm serve LiquidAI/LFM2.5-1.2B-Base \\\n  --tensor-parallel-size 1",
    "argv": [
      "vllm",
      "serve",
      "LiquidAI/LFM2.5-1.2B-Base",
      "--tensor-parallel-size",
      "1"
    ],
    "docker_command": "docker run --gpus all \\\n  --privileged --ipc=host -p 8000:8000 \\\n  -v ~/.cache/huggingface:/root/.cache/huggingface \\\n  vllm/vllm-openai:latest LiquidAI/LFM2.5-1.2B-Base \\\n  --tensor-parallel-size 1",
    "docker_argv": [
      "docker",
      "run",
      "--gpus",
      "all",
      "--privileged",
      "--ipc=host",
      "-p",
      "8000:8000",
      "-v",
      "~/.cache/huggingface:/root/.cache/huggingface",
      "vllm/vllm-openai:latest",
      "LiquidAI/LFM2.5-1.2B-Base",
      "--tensor-parallel-size",
      "1"
    ],
    "strategy_spec": {
      "name": "single_node_tp",
      "deploy_type": "single_node",
      "display_name": "Tensor Parallel",
      "short_label": "TP",
      "orientation": "latency",
      "description": "Single-node tensor parallel. Splits the model across all local GPUs. TP size is set to the GPU count at deploy time. The simplest multi-GPU strategy — works for all model architectures.\n",
      "hardware_match": {
        "min_gpus": 1,
        "max_gpus": 8,
        "multi_node": false
      },
      "vllm_args": [],
      "parallel_flag": "--tensor-parallel-size"
    },
    "hardware_profile": {
      "brand": "NVIDIA",
      "generation": "hopper",
      "compute_arch": "sm90",
      "display_name": "H200",
      "description": "NVIDIA H200 SXM 141 GB HBM3e · 8-GPU HGX node",
      "gpu_count": 8,
      "vram_gb": 1128,
      "multi_node": false
    },
    "alternatives": {
      "kv_store_centralized_mooncake": "/LiquidAI/LFM2.5-1.2B-Base/strategies/kv_store_centralized_mooncake.json",
      "kv_store_distributed_mooncake": "/LiquidAI/LFM2.5-1.2B-Base/strategies/kv_store_distributed_mooncake.json"
    },
    "by_hardware": {
      "h200": "/LiquidAI/LFM2.5-1.2B-Base/hw/h200.json",
      "h100": "/LiquidAI/LFM2.5-1.2B-Base/hw/h100.json",
      "b200": "/LiquidAI/LFM2.5-1.2B-Base/hw/b200.json",
      "gb200": "/LiquidAI/LFM2.5-1.2B-Base/hw/gb200.json",
      "b300": "/LiquidAI/LFM2.5-1.2B-Base/hw/b300.json",
      "gb300": "/LiquidAI/LFM2.5-1.2B-Base/hw/gb300.json",
      "mi300x": "/LiquidAI/LFM2.5-1.2B-Base/hw/mi300x.json",
      "mi325x": "/LiquidAI/LFM2.5-1.2B-Base/hw/mi325x.json",
      "mi350x": "/LiquidAI/LFM2.5-1.2B-Base/hw/mi350x.json",
      "mi355x": "/LiquidAI/LFM2.5-1.2B-Base/hw/mi355x.json"
    }
  },
  "model": {
    "model_id": "LiquidAI/LFM2.5-1.2B-Base",
    "min_vllm_version": "0.23.0",
    "architecture": "dense",
    "parameter_count": "1.2B",
    "active_parameters": "1.2B",
    "context_length": 32768,
    "base_args": [],
    "base_env": {},
    "install": {
      "pip": {
        "command": "uv venv\nsource .venv/bin/activate\nuv pip install -U vllm --torch-backend auto"
      },
      "docker": {
        "command": "docker pull vllm/vllm-openai:latest"
      }
    }
  },
  "variants": {
    "default": {
      "precision": "bf16",
      "vram_minimum_gb": 3,
      "description": "Full BF16 — single 8 GB+ NVIDIA GPU with room for KV cache"
    }
  },
  "hardware_overrides": {
    "blackwell": {
      "extra_env": {
        "VLLM_USE_OINK_OPS": "1"
      }
    }
  },
  "guide": "## Overview\n\n[LFM2.5-1.2B-Base](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Base) is the pretrained base\ncheckpoint behind [Liquid AI's](https://www.liquid.ai/) LFM2.5-1.2B instruct models, built on\nthe **LFM2 hybrid backbone** (short-range gated convolution blocks interleaved with grouped-query\nattention). It is **not** instruction-tuned and has no chat template — use the\n`/v1/completions` endpoint (raw text continuation), or fine-tune it for your own task.\n\n### Key Features\n- **Pretrained base**: No chat template, no instruction tuning — a clean foundation for fine-tuning or raw completion.\n- **Hybrid backbone**: Gated short convolutions interleaved with grouped-query attention — a smaller KV cache and lower decode latency than a same-size full-attention transformer.\n- **32K context**: 32,768-token context window.\n- **Native vLLM support**: Served via the `Lfm2ForCausalLM` architecture.\n\n### Supported Variants\n\nDense:\n- `LiquidAI/LFM2.5-230M` (230M)\n- `LiquidAI/LFM2.5-350M` (350M)\n- `LiquidAI/LFM2.5-1.2B-Instruct` (1.2B)\n- `LiquidAI/LFM2.5-1.2B-Thinking` (1.2B, reasoning)\n- `LiquidAI/LFM2.5-1.2B-JP` / `LiquidAI/LFM2.5-1.2B-JP-202606` (Japanese)\n- `LiquidAI/LFM2.5-1.2B-Base` (pretrained base)\n\nMoE:\n- `LiquidAI/LFM2.5-8B-A1B` (8B total / ~1B active)\n\nVision-Language:\n- `LiquidAI/LFM2.5-VL-450M`, `LiquidAI/LFM2.5-VL-1.6B`\n\nSee the [LFM2.5 usage guide](../../LiquidAI/LFM2.5) for the full family.\n\n## Prerequisites\n\n- **Hardware**: 1× GPU with ≥8 GB VRAM.\n- **vLLM**: ≥ 0.23.0.\n\n### pip (NVIDIA CUDA)\n```bash\nuv venv\nsource .venv/bin/activate\nuv pip install -U vllm --torch-backend auto\n```\n\n## Deployment Configurations\n\n### Quick Start (Single GPU, BF16)\n```bash\nvllm serve LiquidAI/LFM2.5-1.2B-Base\n```\n\n### Docker (NVIDIA)\n```bash\ndocker run -itd --name lfm2.5-base \\\n  --ipc=host --network host --shm-size 16G --gpus all \\\n  -v ~/.cache/huggingface:/root/.cache/huggingface \\\n  vllm/vllm-openai:latest \\\n    --model LiquidAI/LFM2.5-1.2B-Base \\\n    --host 0.0.0.0 --port 8000\n```\n\n## Client Usage\n\n### Text Completion\nThis is a base model — use the **completions** endpoint, not chat. Recommended sampling:\ntemperature `0.3`, `min_p 0.15`, `repetition_penalty 1.05`.\n```python\nfrom openai import OpenAI\nclient = OpenAI(base_url=\"http://localhost:8000/v1\", api_key=\"EMPTY\")\nresponse = client.completions.create(\n    model=\"LiquidAI/LFM2.5-1.2B-Base\",\n    prompt=\"The three laws of thermodynamics are:\",\n    temperature=0.3,\n    extra_body={\"min_p\": 0.15, \"repetition_penalty\": 1.05},\n)\nprint(response.choices[0].text)\n```\n\n## Configuration Tips\n- Use `/v1/completions` (raw continuation) — there is no chat template on a base model.\n- For fine-tuning, this checkpoint is the recommended starting point for custom LFM2.5-1.2B tasks.\n- Set `--max-model-len` to match your workload (up to 32K).\n- Sampling presets are per-request client defaults — don't bake them into `vllm serve`.\n\n## References\n- [Model card](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Base)\n- [LFM2.5 usage guide](../../LiquidAI/LFM2.5)\n- [Liquid AI](https://www.liquid.ai/)\n- [vLLM supported models](https://docs.vllm.ai/en/latest/models/supported_models.html)\n",
  "source": {
    "source": "vllm",
    "repo": "vllm-project/vllm",
    "ref": "main",
    "sha": "389ffebe555705c85337025d5df108d0384d365e",
    "path": "recipes/models/LiquidAI/LFM2.5-1.2B-Base.yaml",
    "edit_url": "https://github.com/vllm-project/vllm/edit/main/recipes/models/LiquidAI/LFM2.5-1.2B-Base.yaml",
    "overlays": []
  }
}