Model Family Catalog
Overview
This page is the human-readable rendering of
contracts/model_family_catalog.json.
Use it to answer three distinct questions without weakening the public meaning
of the support matrix:
- What is currently supported as a public lane?
- What families are implemented in code but not publicly supported?
- What families or capabilities should be added next?
Support Tier vs Coverage State
| Term | Meaning | Source of truth |
|---|
support tier | Public support/assurance posture for a declared lane. Values stay aligned with support_matrix.json. | contracts/support_matrix.json |
coverage state | Repo implementation maturity outside the public support matrix, such as profile_first_class, profile_shared_alias, auto_or_loader_only, loader_only, or backlog states. | contracts/model_family_catalog.json |
The support matrix remains strict. The model family catalog is broader by
design and records code-level visibility, usage-only checkpoints, and
recommended additions. Access-gated vendor checkpoints are intentionally kept
out of declared support lanes and shipped preset inventory.
Declared Support
| Family | State | Representative models | Notes |
|---|
| GPT-2 causal LM | published_basis | openai-community/gpt2 | Public lane derived from gpt2-causal-hf. |
| BERT / RoBERTa MLM | published_basis | bert-base-uncased, roberta-base | Public lane derived from bert-mlm-hf. |
| Mistral 7B causal LM | supported_experimental | mistralai/Mistral-7B-v0.1 | Pilot preset and calibration config are shipped. |
| Qwen2 7B causal LM | supported_experimental | Qwen/Qwen2-7B | Pilot preset and calibration config are shipped. |
| Qwen3 causal LM | supported_experimental | Qwen/Qwen3-8B | Pilot preset and calibration config are shipped. |
| QwQ-32B reasoning causal LM | supported_experimental | Qwen/QwQ-32B | Pilot preset and calibration config are shipped, and the current remote evaluate/verify lane closes cleanly on the dense checkpoint. |
| DeepSeek-R1-Distill-Qwen causal LM | supported_experimental | deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | Pilot preset and calibration config are shipped. |
| Phi-4 causal LM (text-only eval) | supported_experimental | microsoft/Phi-4-reasoning-plus | Text-only pilot preset and calibration config are shipped, and the current remote evaluate/verify lane closes cleanly. |
| TinyLlama 1.1B causal LM | supported_experimental | TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Ungated Llama-family pilot lane with shipped preset and calibration config. |
| OLMo 2 causal LM | supported_experimental | allenai/OLMo-2-1124-7B, allenai/OLMo-2-1124-13B-Instruct | Pilot presets and calibration configs are shipped for both 7B and 13B scale points. |
| Qwen3.5 causal LM | supported_experimental | Qwen/Qwen3.5-9B | Pilot preset and calibration config are shipped. |
| Seq2Seq / local pairs | community_experimental | t5-small, facebook/bart-base | Generic seq2seq lane without a published-basis claim. |
Implemented Coverage
| Family | Coverage state | Representative models | Notes |
|---|
| Mixtral | profile_first_class | mistralai/Mixtral-8x7B-v0.1 | Profile and loader code recognize the family directly. |
| Llama | profile_first_class | openlm-research/open_llama_7b, TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Generic Llama-family profile handling is first-class. TinyLlama now provides the ungated declared support lane, while access-gated vendor checkpoints remain omitted. |
| Qwen family aliases (Qwen1.5/Qwen2.5/Qwen3 naming) | profile_first_class | Qwen/Qwen2.5-14B, Qwen/Qwen3.5-9B, Qwen/QwQ-32B | Shared qwen-family heuristics still cover aliases beyond the declared Qwen2, Qwen3, and Qwen3.5 lanes, including the ungated QwQ reasoning branch. |
| Yi | profile_first_class | 01-ai/Yi-34B | Treated as a RoPE decoder family in profile logic. |
| Phi family | profile_first_class | microsoft/Phi-3-mini-4k-instruct, microsoft/Phi-4-reasoning-plus | Dedicated phi-family selectors now exist. Phi-4 now has a declared text-only lane, while multimodal Phi-4 remains backlog-only. |
| OPT / GPT-NeoX / GPT-J | profile_shared_alias | facebook/opt-1.3b, EleutherAI/gpt-neox-20b | Available through shared GPT-style paths. |
| Falcon | auto_or_loader_only | tiiuae/falcon-7b | Visible through adapter-auto heuristics only. |
| GLM | auto_or_loader_only | THUDM/glm-4-9b-chat | Visible through adapter-auto heuristics only. |
| DeepSeek | profile_first_class | deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | DeepSeek distill checkpoints continue to share the qwen-family route. Oversized FP8 checkpoint-specific repo hooks and shipped configs were removed after bring-up showed that they do not fit the supported hardware/runtime path. |
| Broader BERT-like MLMs (DistilBERT/ALBERT/DeBERTa/ELECTRA) | auto_or_loader_only | distilbert-base-uncased, microsoft/deberta-v3-base | Loader/auto support exceeds the public BERT / RoBERTa lane. |
| Broader seq2seq families (mBART/PEGASUS/Marian) | auto_or_loader_only | facebook/mbart-large-50, Helsinki-NLP/opus-mt-en-de | Loader support is broader than the generic seq2seq public lane. |
Usage Only
| Family | State | Representative models | Notes |
|---|
| Qwen2.5 family | usage_only | Qwen/Qwen2.5-7B, Qwen/Qwen2.5-14B, Qwen/Qwen2.5-32B | Used in proof-pack suites and validation defaults. |
| Qwen1.5 72B | usage_only | Qwen/Qwen1.5-72B | Used concretely in proof-pack suites. |
| Yi 34B | usage_only | 01-ai/Yi-34B | Used in workshop and full proof-pack suites. |
| Mixtral 8x7B | usage_only | mistralai/Mixtral-8x7B-v0.1 | Used in proof-pack flows without a public support lane. |
Recommended Additions
| Priority | Family | Planned support mode | Representative models | Notes |
|---|
P2 | Full multimodal evaluation pipeline | full_multimodal_eval | microsoft/Phi-4-vision-reasoning-15B | Deferred capability backlog item beyond text-only evaluation for ungated multimodal checkpoints. |
A family only moves into support_matrix.json after all of the following are
present:
- explicit adapter/profile recognition
- a shipped preset
- a shipped calibration config
- targeted tests
- CLI smoke evidence
- approved calibration/evaluation evidence