Skip to content

fix(admin): clarify DFlash draft quantization option + FP16 draft model boost #880

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
deepsweet wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(admin): clarify DFlash draft quantization option + FP16 draft model boost #880

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions docs/experimental/dflash_mlx_integration.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,7 +90,7 @@ Other model families (Llama, Gemma, etc.) are not supported — they require bot
|---------|------|-------------|
| `dflash_enabled` | bool | Enable/disable DFlash for this model |
| `dflash_draft_model` | str | Path or HuggingFace repo for draft checkpoint |
| `dflash_draft_quant_bits` | int\|None | Draft model quantization (None=bf16, 4=int4) |
| `dflash_draft_quant_bits` | int\|None | Draft model quantization (None=bf16/fp16, 4=int4) |

Configured via web admin UI → Model Settings → Experimental Features → DFlash.

Expand Down Expand Up @@ -240,7 +240,7 @@ DFlash check runs **before** engine type routing in `_load_engine()`. If `dflash
Located in Model Settings → Advanced Settings → Experimental Features → DFlash:
- **Toggle**: enable/disable DFlash
- **Draft Model**: dropdown of available models
- **Draft Quantization**: bf16 (default) / 4-bit
- **Draft Quantization**: bf16/fp16 (default) / 4-bit

### Logging

Expand Down
3 changes: 1 addition & 2 deletions omlx/admin/templates/dashboard/_modal_model_settings.html
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -633,9 +633,8 @@ <h4 class="text-xs font-bold uppercase tracking-widest text-neutral-400 mb-3">{{
<label class="block text-xs font-bold uppercase tracking-wider text-neutral-500 mb-2">Draft Quantization</label>
<select x-model="modelSettings.dflash_draft_quant_bits"
class="w-full px-4 py-2.5 border border-neutral-200 rounded-xl text-sm focus:ring-2 focus:ring-neutral-900 focus:border-transparent transition-all bg-white">
<option value="">bf16 (default)</option>
<option value="">None (default)</option>
<option value="4">4-bit</option>
<option value="8">8-bit</option>
</select>
</div>
</div>
Expand Down
2 changes: 1 addition & 1 deletion omlx/model_settings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,7 +107,7 @@ class ModelSettings:
# DFlash (block diffusion speculative decoding)
dflash_enabled: bool = False
dflash_draft_model: Optional[str] = None # Path/repo for DFlash draft checkpoint
dflash_draft_quant_bits: Optional[int] = None # Draft model quantization (None=bf16, 4)
dflash_draft_quant_bits: Optional[int] = None # Draft model quantization (None=bf16/fp16, 4)

# Model management flags
is_pinned: bool = False
Expand Down