Instructions to use Youssofal/Qwen3.6-27B-MTPLX-Optimized-Speed-FP16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Youssofal/Qwen3.6-27B-MTPLX-Optimized-Speed-FP16 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Qwen3.6-27B-MTPLX-Optimized-Speed-FP16 Youssofal/Qwen3.6-27B-MTPLX-Optimized-Speed-FP16
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
Clean release metadata
Browse files- mtplx_runtime.json +4 -28
mtplx_runtime.json
CHANGED
|
@@ -15,7 +15,7 @@
|
|
| 15 |
"verify_tokens": 4
|
| 16 |
},
|
| 17 |
"mtp_depth_max": 3,
|
| 18 |
-
"mtp_sidecar": "
|
| 19 |
"mtplx_version": "0.1.0-preview",
|
| 20 |
"precision_policy": {
|
| 21 |
"intended_default_for": [
|
|
@@ -24,7 +24,6 @@
|
|
| 24 |
],
|
| 25 |
"note": "This is a sibling precision variant; it is not a universal speed claim.",
|
| 26 |
"routing": "mtplx start auto-selects this artifact on M1/M2 Apple Silicon",
|
| 27 |
-
"source_repo": "Youssofal/Qwen3.6-27B-MTPLX-Optimized-Speed",
|
| 28 |
"variant": "fp16"
|
| 29 |
},
|
| 30 |
"precision_variant": "fp16",
|
|
@@ -38,7 +37,7 @@
|
|
| 38 |
"top_k": 20,
|
| 39 |
"top_p": 0.95
|
| 40 |
},
|
| 41 |
-
"recommended_profile": "
|
| 42 |
"sampler": {
|
| 43 |
"temperature": 0.6,
|
| 44 |
"top_k": 20,
|
|
@@ -55,10 +54,6 @@
|
|
| 55 |
48,
|
| 56 |
46
|
| 57 |
],
|
| 58 |
-
"artifacts": [
|
| 59 |
-
"/Users/youssof/.mtplx/fanmax_flat4_draft_temp070_cleanapps_20260503-230229.json",
|
| 60 |
-
"/Users/youssof/.mtplx/fanmax_flat4_draft_temp070_cleanapps_repeat_20260503-230458.json"
|
| 61 |
-
],
|
| 62 |
"correction_tokens": 3,
|
| 63 |
"depth": 3,
|
| 64 |
"draft_lm_head": "3-bit affine group64 draft-only head",
|
|
@@ -79,7 +74,6 @@
|
|
| 79 |
49,
|
| 80 |
43
|
| 81 |
],
|
| 82 |
-
"artifact": "/Users/youssof/.mtplx/fanmax_flat4_full_greedy_cleanapps_20260503-230229.json",
|
| 83 |
"correction_tokens": 0,
|
| 84 |
"draft_sampler": {
|
| 85 |
"temperature": 0.0,
|
|
@@ -99,33 +93,15 @@
|
|
| 99 |
"tok_s": 60.108346853627076,
|
| 100 |
"verify_calls": 51
|
| 101 |
},
|
| 102 |
-
"older_3bit_artifacts": [
|
| 103 |
-
"/Users/youssof/.mtplx/fanmax_flat4_cyankiwi_draft_head_3bit_isolated_20260503.json",
|
| 104 |
-
"/Users/youssof/.mtplx/fanmax_flat4_cyankiwi_draft_head_3bit_repeat_20260503.json"
|
| 105 |
-
],
|
| 106 |
"older_3bit_tok_s": [
|
| 107 |
60.03822277200243,
|
| 108 |
60.06063543439426
|
| 109 |
],
|
| 110 |
-
"previous_best_artifacts": [
|
| 111 |
-
"/Users/youssof/.mtplx/fanmax_flat4_draft_temp070_optimized_env_20260503.json",
|
| 112 |
-
"/Users/youssof/.mtplx/fanmax_flat4_draft_temp070_repeat_optimized_env_20260503.json"
|
| 113 |
-
],
|
| 114 |
"previous_best_tok_s": [
|
| 115 |
61.52744855840259,
|
| 116 |
60.89966247416089
|
| 117 |
],
|
| 118 |
-
"
|
| 119 |
-
{
|
| 120 |
-
"artifact": "/Users/youssof/.mtplx/fanmax_mtplx_flat4_cyankiwi_cli_contract_perfcold_disablethinking_20260503.json",
|
| 121 |
-
"tok_s": 59.51602930081416
|
| 122 |
-
},
|
| 123 |
-
{
|
| 124 |
-
"artifact": "/Users/youssof/.mtplx/fanmax_mtplx_flat4_cyankiwi_cli_contract_perfcold_disablethinking_repeat_20260503.json",
|
| 125 |
-
"tok_s": 59.97165246907193
|
| 126 |
-
}
|
| 127 |
-
],
|
| 128 |
-
"profile": "performance-cold",
|
| 129 |
"timestamp": "2026-05-03T23:07:00+0100",
|
| 130 |
"tok_s": [
|
| 131 |
63.05651084048216,
|
|
@@ -136,7 +112,7 @@
|
|
| 136 |
"verified_on": {
|
| 137 |
"hardware": "Apple M5 Max, 128 GB unified memory",
|
| 138 |
"machine_arch": "arm64",
|
| 139 |
-
"model": "Qwen3.6-27B-MTPLX-Flat4-
|
| 140 |
"timestamp": "2026-05-03T17:49:00+0100"
|
| 141 |
}
|
| 142 |
}
|
|
|
|
| 15 |
"verify_tokens": 4
|
| 16 |
},
|
| 17 |
"mtp_depth_max": 3,
|
| 18 |
+
"mtp_sidecar": "native MTP sidecar",
|
| 19 |
"mtplx_version": "0.1.0-preview",
|
| 20 |
"precision_policy": {
|
| 21 |
"intended_default_for": [
|
|
|
|
| 24 |
],
|
| 25 |
"note": "This is a sibling precision variant; it is not a universal speed claim.",
|
| 26 |
"routing": "mtplx start auto-selects this artifact on M1/M2 Apple Silicon",
|
|
|
|
| 27 |
"variant": "fp16"
|
| 28 |
},
|
| 29 |
"precision_variant": "fp16",
|
|
|
|
| 37 |
"top_k": 20,
|
| 38 |
"top_p": 0.95
|
| 39 |
},
|
| 40 |
+
"recommended_profile": "sustained",
|
| 41 |
"sampler": {
|
| 42 |
"temperature": 0.6,
|
| 43 |
"top_k": 20,
|
|
|
|
| 54 |
48,
|
| 55 |
46
|
| 56 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
"correction_tokens": 3,
|
| 58 |
"depth": 3,
|
| 59 |
"draft_lm_head": "3-bit affine group64 draft-only head",
|
|
|
|
| 74 |
49,
|
| 75 |
43
|
| 76 |
],
|
|
|
|
| 77 |
"correction_tokens": 0,
|
| 78 |
"draft_sampler": {
|
| 79 |
"temperature": 0.0,
|
|
|
|
| 93 |
"tok_s": 60.108346853627076,
|
| 94 |
"verify_calls": 51
|
| 95 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
"older_3bit_tok_s": [
|
| 97 |
60.03822277200243,
|
| 98 |
60.06063543439426
|
| 99 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
"previous_best_tok_s": [
|
| 101 |
61.52744855840259,
|
| 102 |
60.89966247416089
|
| 103 |
],
|
| 104 |
+
"profile": "sustained",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 105 |
"timestamp": "2026-05-03T23:07:00+0100",
|
| 106 |
"tok_s": [
|
| 107 |
63.05651084048216,
|
|
|
|
| 112 |
"verified_on": {
|
| 113 |
"hardware": "Apple M5 Max, 128 GB unified memory",
|
| 114 |
"machine_arch": "arm64",
|
| 115 |
+
"model": "Qwen3.6-27B-MTPLX-Flat4-calibratedMTP",
|
| 116 |
"timestamp": "2026-05-03T17:49:00+0100"
|
| 117 |
}
|
| 118 |
}
|