Skip to main content

Architecture & Backend Compatibility

This is the single reference for which model architectures ColabHive can serve and on which inference backends. Architecture support is 100% database-driven: when you register a model, the platform reads its config.json architecture and picks a compatible backend automatically.

Check your specific model, not this list

Counts and per-architecture details are a snapshot. To know whether a specific HuggingFace repo will run — and on which backend — ask the live compatibility check, which reads the repo and returns an authoritative answer:

curl "https://api.colabhive.com/api/builder/v1/models/hf/{owner}/{repo}/info" \
-H "X-API-Key: hive_..."

The compatibility object reports status (compatible · requires_review · incompatible), the runtime_profile_id (backend that will be used), and checks; estimated_requirements reports VRAM and a recommended config.


How a backend is chosen

ColabHive selects an inference backend from the model's architecture:

  • vLLM (native). Architectures with first-class vLLM support get PagedAttention, continuous batching, and CUDA graphs — the fast path for production LLM serving.
  • Transformers (universal fallback). Any HuggingFace *ForCausalLM (and custom modeling_*.py architectures) that vLLM does not natively accelerate falls back to the transformers backend. Slower, but 100% compatible.
  • Specialized backends. Diffusion/generative models, time-series specialists, SSM/hybrid models, and CPU GGUF (llama.cpp) each route to their own backend (see the backend table below).

You do not choose the backend manually — registration resolves it from the architecture and the model's runtime profile.


Supported architectures (snapshot 2026-08-15)

As of this snapshot the registry recognizes 393 active architectures across the families below. This is a moving target — it more than quadrupled between the 2026-07-23 and 2026-08-15 snapshots; verify a specific repo with the compatibility check above.

Representative families (not exhaustive):

FamilyExample architecturesExample models
LlamaLlamaForCausalLM, Llama4ForCausalLM, CodeLlamaForCausalLMLlama 2/3/4, CodeLlama, Vicuna
MistralMistralForCausalLM, MixtralForCausalLMMistral 7B, Mixtral MoE
QwenQwen2ForCausalLM, Qwen3ForCausalLM, Qwen2MoeForCausalLM, QWenLMHeadModelQwen 1/2.5/3, Qwen MoE
GemmaGemmaForCausalLM, Gemma2ForCausalLM, Gemma3ForCausalLMGemma, Gemma 2, Gemma 3
PhiPhiForCausalLM, Phi3ForCausalLM, PhiMoEForCausalLMPhi-2, Phi-3/3.5, Phi MoE
DeepSeekDeepseekForCausalLM, DeepseekV2ForCausalLM, DeepseekV3ForCausalLMDeepSeek, DeepSeek Coder, v2/v3
GPT / classicGPT2LMHeadModel, GPTNeoXForCausalLM, GPTJForCausalLM, BloomForCausalLM, OPTForCausalLM, FalconForCausalLM, MPTForCausalLMGPT-2, Pythia, GPT-J, BLOOM, OPT, Falcon, MPT
GLM / ChatGLMChatGLMModel, Glm4ForCausalLMChatGLM, GLM-4
MoE (incl. MLA)Glm4MoeForCausalLM, Qwen3MoeForCausalLM, GptOssForCausalLM, NemotronHForCausalLMGLM-4.7 Flash, Qwen3-Coder 30B-A3B, gpt-oss, Nemotron-3 Super
Multimodal / vision-languageQwen3VLForConditionalGeneration, Qwen3VLMoeForConditionalGeneration, Mistral3ForConditionalGeneration, Glm4vForConditionalGeneration, InternVLForConditionalGeneration, KimiVLForConditionalGeneration, NemotronH_Nano_VL_V2Qwen3-VL, Ministral 3, GLM-4V, InternVL, Kimi-VL, Nemotron Nano VL
Other LLMsBaichuanForCausalLM, InternLMForCausalLM, InternLM2ForCausalLM, YiForCausalLM, XverseForCausalLM, CohereForCausalLM, Cohere2ForCausalLM, DbrxForCausalLM, ArcticForCausalLM, NemotronForCausalLM, GraniteForCausalLM, OlmoForCausalLM, Olmo2ForCausalLM, StableLmForCausalLM, Starcoder2ForCausalLM, ExaoneForCausalLM, SolarForCausalLM, …InternLM, Yi, Command, DBRX, Arctic, Nemotron, Granite, OLMo, StableLM, StarCoder 2, EXAONE, SOLAR
SSM / hybridMambaForCausalLM, Mamba2ForCausalLM, FalconMambaForCausalLM, JambaForCausalLM, ZambaForCausalLMMamba, Falcon-Mamba, Jamba, Zamba — fine-tuning templates only (see below)
Diffusion / generativediffusers pipelinesSDXL, SD3, FLUX, MusicGen, AudioLDM2 (see Generative)
Time seriesPatchTSTForPrediction, PatchTSTForClassification, BiTCN, TimesFMPatchTST, BiTCN, TimesFM 2.5
SSM / hybrid (Mamba, Jamba, Zamba) are training templates, not inference endpoints

These architectures are present for fine-tuning (mamba-ssm-finetune, jamba-hybrid-finetune). They are not benchmarked and there are no public SSM inference endpoints. In the compatibility data their rows are untested (throughput unknown). Do not treat them as ready-to-call chat models.


Benchmarked throughput (tested)

Only a small set of architecture/backend pairs is marked tested (14 of them today), and only the ten below carry a real throughput number — all on the vllm-cu121 (NVIDIA) backend. All other rows in the compatibility table are untested with no throughput figure — treat those as "should work", not "measured".

ArchitectureBackendThroughput (tok/s)Benchmark model
GPT-2 (GPT2LMHeadModel)vllm-cu121150gpt2
Mixtral MoE (MixtralForCausalLM)vllm-cu121120mistralai/Mixtral-8x7B-Instruct-v0.1
Phi-2 (PhiForCausalLM)vllm-cu121110microsoft/phi-2
Phi-3 (Phi3ForCausalLM)vllm-cu121105microsoft/Phi-3-mini-4k-instruct
Llama (LlamaForCausalLM)vllm-cu121100meta-llama/Llama-2-7b-chat-hf
Mistral (MistralForCausalLM)vllm-cu12195mistralai/Mistral-7B-Instruct-v0.3
Gemma (GemmaForCausalLM)vllm-cu12190google/gemma-7b-it
Qwen 2 (Qwen2ForCausalLM)vllm-cu12190Qwen/Qwen2.5-7B-Instruct
Qwen 1 (QWenLMHeadModel)vllm-cu12185Qwen/Qwen-7B-Chat
GPT-NeoX (GPTNeoXForCausalLM)vllm-cu12180EleutherAI/pythia-6.9b

Numbers are single-GPU order-of-magnitude figures for the benchmark model, not a guarantee for every model in the family. Transformers-fallback serving is typically several times slower than native vLLM.

The four remaining tested pairs are the GLM-4 MoE architectures (Glm4MoeForCausalLM, Glm4MoeLiteForCausalLM and their MTP variants) on vllm-xpu — validated on Intel Arc on 2026-08-03 during the GLM-4.7 Flash bring-up, recorded as tested but without a throughput figure.


Inference backends

ColabHive runs 16 active inference backends across NVIDIA, AMD, Intel, and CPU — not just vLLM. Thirteen of them appear in the current compatibility matrix. Images are referenced by repo; the platform resolves the running version from system_versions at dispatch (never :latest), with two legacy rows still carrying an explicit v1.0.0 pin.

backend_idVendorFrameworkImage (repo)
vllm-cu121NVIDIAvLLMregistry.colabhive.com/inference-vllm
transformers-cu121NVIDIAtransformersregistry.colabhive.com/inference-base
tgi-cu121NVIDIATGIregistry.colabhive.com/inference-tgi:v1.0.0
mamba-vllm-cu121NVIDIAvLLM (+mamba-ssm)registry.colabhive.com/inference-vllm
diffusers-cu121NVIDIAdiffusersregistry.colabhive.com/inference-generative
generative-cu121NVIDIAtransformersregistry.colabhive.com/inference-generative
specialists-ts-cu121NVIDIAspecialistsregistry.colabhive.com/inference-specialists
vllm-rocmAMDvLLMregistry.colabhive.com/inference-rocm
transformers-rocmAMDtransformersregistry.colabhive.com/inference-rocm
specialists-rocmAMDspecialistsregistry.colabhive.com/inference-rocm
vllm-xpuIntelvLLMregistry.colabhive.com/inference-ipex
ipex-xeInteltransformers (IPEX)registry.colabhive.com/inference-ipex
specialists-ipex-xpuIntelspecialistsregistry.colabhive.com/inference-ipex
transformers-cpuCPUtransformersregistry.colabhive.com/inference-base
specialist-cpu-optimizedCPUspecialistsregistry.colabhive.com/inference-base-cpu:v1.0.0
llm-llamacpp-cpuCPUllama.cppregistry.colabhive.com/inference-llamacpp-cpu

Intel XPU (vllm-xpu / ipex-xe) and AMD ROCm backends are in production alongside NVIDIA — each vendor now has a matching specialists backend as well, so embeddings/rerank/OCR/STT/moderation run on whatever silicon the node has. The CPU GGUF backend (llm-llamacpp-cpu) powers the CPU LLM tier.

Superseded backend IDs

vllm-rocm60 and transformers-rocm60 were the earlier AMD entries pointing at a local inference-base-rocm image. They are deactivated — the live AMD backends are vllm-rocm, transformers-rocm and specialists-rocm on registry.colabhive.com/inference-rocm.


Requesting or adding architecture support

  1. Try the compatibility check first — most models resolve to vLLM-native or the transformers fallback automatically.
  2. If a model reports incompatible or you need native vLLM acceleration for a new architecture, open a GitHub issue with the repo ID, the architecture name from config.json, and your use case.


Authors: José Luis Minich, Maximiliano Lucius.