{{ t('models.manager.description') }}
{{ t('models.manager.no_models') }}
{{ t('models.downloader.description') }}
{{ t('models.browse.loading') }}
{{ t('models.browse.load_prompt') }}
|
{{ t('models.browse.tight_fit') }}
{{ t('models.browse.high_oom_risk') }}
|
{{ t('models.browse.tight_fit') }}
{{ t('models.browse.high_oom_risk') }}
|
||||
{{ t('models.browse.searching') }}
{{ t('models.browse.search_prompt') }}
{{ t('models.browse.no_results') }}
|
{{ t('models.browse.tight_fit') }}
{{ t('models.browse.high_oom_risk') }}
|
{{ t('models.browse.tight_fit') }}
{{ t('models.browse.high_oom_risk') }}
|
{{ t('models.browse.loading') }}
{{ t('models.browse.load_prompt') }}
{{ t('models.browse.no_trending') }}
{{ t('models.browse.no_popular') }}
{{ t('models.browse.searching') }}
{{ t('models.browse.search_prompt') }}
{{ t('models.browse.no_results') }}
{{ t('models.oq.description') }}
{{ t('models.oq.enhanced_help') }}
{{ t('models.oq.imatrix_reuse_cache_help') }}
{{ t('models.oq.imatrix_strict_help') }}
Excludes vision encoder weights. Output is a text-only model (~2-3% smaller).
{{ t('models.oq.preserve_mtp_help') }}
{{ t('models.oq.preserve_mtp_unavailable') }}
{{ t('models.oq.dtype_help') }}
Quantization should not be exclusive to any particular inference server.
oQ produces standard mlx-lm models that work everywhere — oMLX, mlx-lm, LM Studio, and any app that supports MLX safetensors format.
No custom loader required.
The quantizer streams tensors from safetensors, measures layer sensitivity, and writes a byte-budgeted mixed-precision checkpoint. Standard oQ focuses on layer-level sensitivity and model-aware protection rules. oQe keeps that same oQ plan and adds imatrix-weighted affine quantization.
oQe uses the same oQ sensitivity planner, then adds an importance-matrix calibration step. This is explicitly borrowed from the llama.cpp imatrix idea: collect activation energy from representative prompts, then use that information so quantization spends less error on the input channels that matter most.
| Path | Bit allocation | Affine quantization | Best use |
|---|---|---|---|
| oQ | Layer-sensitivity mixed precision | Standard min/max affine per group | Fast, deterministic conversion with strong baseline quality |
| oQe | Same oQ plan | imatrix-weighted clipping/search per group | Better low-bit retention, especially for MoE and activation-skewed layers |
If a tensor has no matching imatrix entry, the default behavior is to fall back to standard oQ for that tensor. Enable strict coverage to fail instead.
{{ t('models.detail.loading') }}
{{ t('models.detail.no_model_card') }}
{{ t('models.detail.no_files') }}
{{ t('models.detail.no_tags') }}