Split modelsVERSION 1 Starter template

SmolLM2 split endpoints template

An unbenchmarked split starting point with SmolLM2 135M and 360M on separate endpoints. Community-authored CC0 recipe content; model weights retain their own upstream licenses.

TwinSpark Community·Oct 29, 2026

Download JSON

A starting point, ready for your own checks. This template has not been benchmarked. Review the configuration and validate it on your hardware before running.

IMPORT THE LATEST VERSIONhttps://studio-tools.info/api/v1/recipes/smollm2-split-endpoints-template/latest.json

Use this HTTPS URL in TwinSpark Manager. It follows the latest published version.

THE INGREDIENTS

Recipe overview

Primary model: HuggingFaceTB/SmolLM2-135M-Instruct
Secondary model: HuggingFaceTB/SmolLM2-360M-Instruct
Topology: Split models
License: CC0-1.0
Quantization: BF16
Context length: 1,024 tokens
Capabilities: Standard inference

VERSION NOTES

Original TwinSpark Community template. Unbenchmarked; validate hardware compatibility, runtime, and model license before use.

TAKE IT TO YOUR WORKSPACE

The full configuration

{
    "advanced": {
        "attention_backend": null,
        "block_size": null,
        "chunked_prefill": null,
        "compilation_config": null,
        "download_include": [],
        "dtype": null,
        "eager_mode": true,
        "env": [],
        "extra_models": [],
        "extra_vllm_args": [],
        "extra_vllm_flags": [],
        "gpu_memory_utilization": null,
        "head_start_delay_s": 0,
        "kv_dtype": null,
        "max_num_batched_tokens": null,
        "max_num_seqs": null,
        "max_num_seqs_vllm_default": false,
        "mods": [],
        "prefix_cache": true,
        "speculative_config": null,
        "tokenizer_mode": null,
        "trust_remote_code": false,
        "weight_loader": null
    },
    "behaviour": {
        "chat_template": null,
        "chat_template_kwargs": [],
        "manage_thinking_kwarg": true,
        "max_tokens": null,
        "min_p": null,
        "reasoning_parser": null,
        "repetition_penalty": null,
        "temperature": null,
        "tool_call_parser": null,
        "top_k": null,
        "top_p": null
    },
    "description": "Community-authored, unbenchmarked split template with two independent small-model endpoints. No measured performance or verified hardware compatibility is claimed.",
    "distributed_backend": "auto",
    "identity": null,
    "image_hint": null,
    "name": "community-smollm2-split",
    "runtime_adapter": "eugr",
    "secondary": {
        "advanced": {
            "attention_backend": null,
            "block_size": null,
            "chunked_prefill": null,
            "compilation_config": null,
            "download_include": [],
            "dtype": null,
            "eager_mode": true,
            "env": [],
            "extra_models": [],
            "extra_vllm_args": [],
            "extra_vllm_flags": [],
            "gpu_memory_utilization": null,
            "head_start_delay_s": 0,
            "kv_dtype": null,
            "max_num_batched_tokens": null,
            "max_num_seqs": null,
            "max_num_seqs_vllm_default": false,
            "mods": [],
            "prefix_cache": true,
            "speculative_config": null,
            "tokenizer_mode": null,
            "trust_remote_code": false,
            "weight_loader": null
        },
        "behaviour": {
            "chat_template": null,
            "chat_template_kwargs": [],
            "manage_thinking_kwarg": true,
            "max_tokens": null,
            "min_p": null,
            "reasoning_parser": null,
            "repetition_penalty": null,
            "temperature": null,
            "tool_call_parser": null,
            "top_k": null,
            "top_p": null
        },
        "description": "Community-authored, unbenchmarked single-node starting template. No measured performance or verified hardware compatibility is claimed.",
        "distributed_backend": "auto",
        "identity": null,
        "image_hint": null,
        "name": "community-smollm2-secondary",
        "runtime_adapter": "eugr",
        "secondary": null,
        "simple": {
            "api_alias": "community-second",
            "concurrency": 1,
            "context_length": 1024,
            "extra_aliases": [],
            "memory_strategy": "balanced",
            "model": "HuggingFaceTB/SmolLM2-360M-Instruct",
            "quantization": "bf16",
            "thinking": false,
            "tool_calling": false,
            "topology": "single-b"
        },
        "source": [],
        "tags": [
            "community-template",
            "unbenchmarked"
        ],
        "verification": "experimental"
    },
    "simple": {
        "api_alias": "community-small",
        "concurrency": 1,
        "context_length": 1024,
        "extra_aliases": [],
        "memory_strategy": "balanced",
        "model": "HuggingFaceTB/SmolLM2-135M-Instruct",
        "quantization": "bf16",
        "thinking": false,
        "tool_calling": false,
        "topology": "split"
    },
    "source": [],
    "tags": [
        "community-template",
        "unbenchmarked"
    ],
    "verification": "experimental"
}

SHA-256e3506c9f41b7ac337d4d138eb94af667b741aa16754eb4b76190426a82b004a7

EVERY ITERATION HAS A STORY

Version history

Version 1 Viewing
Original TwinSpark Community template. Unbenchmarked; validate hardware compatibility, runtime, and model license before use.
TwinSpark Community · Oct 29, 2026
↗