Instructions to use Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision")
config = load_config("Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Jonandrop/Qwen3.6-27B-MTPLX-Optimized-Speed-Vision" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Jonandrop commited on Jul 18

Commit

7ae8757

verified ·

1 Parent(s): 9f47ef8

fix: honest M5 Pro 64GB speed_evidence + verified_on (committed policy, recommended draft 0.7, warm median). M5 Max historical preserved

Browse files

Files changed (1) hide show

mtplx_runtime.json +146 -50

mtplx_runtime.json CHANGED Viewed

@@ -38,74 +38,170 @@
   },
   "speed_evidence": {
     "acceptance_by_depth": [
-      1.0,
-      0.9795918367346939,
-      0.9387755102040817
     ],
-    "accepted_by_depth": [
-      49,
-      48,
-      46
-    ],
-    "correction_tokens": 3,
     "depth": 3,
-    "draft_lm_head": "3-bit affine group64 draft-only head",
     "draft_sampler": {
       "temperature": 0.7,
       "top_k": 20,
       "top_p": 0.95
     },
-    "drafted_by_depth": [
-      49,
-      49,
-      49
-    ],
     "enable_thinking": false,
-    "greedy_diagnostic": {
       "accepted_by_depth": [
         49,
-        49,
-        43
       ],
-      "correction_tokens": 0,
       "draft_sampler": {
-        "temperature": 0.0,
-        "top_k": 0,
-        "top_p": 1.0
       },
       "drafted_by_depth": [
-        51,
-        51,
-        51
       ],
-      "target_sampler": {
-        "temperature": 0.0,
-        "top_k": 0,
-        "top_p": 1.0
       },
-      "tok_s": 60.108346853627076,
-      "verify_calls": 51
-    },
-    "older_3bit_tok_s": [
-      60.03822277200243,
-      60.06063543439426
-    ],
-    "previous_best_tok_s": [
-      61.52744855840259,
-      60.89966247416089
-    ],
-    "profile": "sustained",
-    "timestamp": "2026-05-03T23:07:00+0100",
-    "tok_s": [
-      63.05651084048216,
-      62.885564723941826
-    ],
-    "verify_calls": 49
   },
   "verified_on": {
-    "hardware": "Apple M5 Max, 128 GB unified memory",
     "machine_arch": "arm64",
-    "model": "Qwen3.6-27B-MTPLX-Flat4-calibratedMTP",
-    "timestamp": "2026-05-03T17:49:00+0100"
   }
 }

   },
   "speed_evidence": {
     "acceptance_by_depth": [
+      0.9,
+      0.7593,
+      0.6476
     ],
     "depth": 3,
+    "verdict": "mtp_depth_wins",
+    "profile": "sustained",
+    "forge_verify_rows": [
+      {
+        "acceptance_by_position": [],
+        "depth": 0,
+        "finish_reasons": {
+          "stop": 1
+        },
+        "hit_token_budget": false,
+        "hit_token_budget_count": 0,
+        "multiplier_vs_ar": 1.0,
+        "quality_passed": true,
+        "tok_s": 13.66,
+        "tok_s_decode": 14.49,
+        "tok_s_e2e": 13.66,
+        "verify_time_s": null
+      },
+      {
+        "acceptance_by_position": [
+          0.9233
+        ],
+        "depth": 1,
+        "finish_reasons": {
+          "stop": 1
+        },
+        "hit_token_budget": false,
+        "hit_token_budget_count": 0,
+        "multiplier_vs_ar": 1.4381,
+        "quality_passed": true,
+        "tok_s": 19.65,
+        "tok_s_decode": 20.9,
+        "tok_s_e2e": 19.65,
+        "verify_time_s": null
+      },
+      {
+        "acceptance_by_position": [
+          0.9496,
+          0.7526
+        ],
+        "depth": 2,
+        "finish_reasons": {
+          "stop": 1
+        },
+        "hit_token_budget": false,
+        "hit_token_budget_count": 0,
+        "multiplier_vs_ar": 1.4594,
+        "quality_passed": true,
+        "tok_s": 19.94,
+        "tok_s_decode": 20.79,
+        "tok_s_e2e": 19.94,
+        "verify_time_s": null
+      },
+      {
+        "acceptance_by_position": [
+          0.9,
+          0.7593,
+          0.6476
+        ],
+        "depth": 3,
+        "finish_reasons": {
+          "stop": 1
+        },
+        "hit_token_budget": false,
+        "hit_token_budget_count": 0,
+        "multiplier_vs_ar": 1.3208,
+        "quality_passed": true,
+        "tok_s": 18.05,
+        "tok_s_decode": 19.71,
+        "tok_s_e2e": 18.05,
+        "verify_time_s": null
+      }
+    ],
+    "greedy_diagnostic": {
+      "tok_s": 13.66
+    },
+    "tok_s": [
+      18.05
+    ],
     "draft_sampler": {
       "temperature": 0.7,
       "top_k": 20,
       "top_p": 0.95
     },
     "enable_thinking": false,
+    "mtp_history_policy": "committed",
+    "measurement_notes": "Re-measured 2026-07-18 on Apple M5 Pro 64GB, MTPLX 1.0.4, mtp_history_policy=committed (CLI default), draft_temperature=0.7 (model recommended_draft_sampler), enable_thinking=false, warm (4 warmup prompts), 8 measured prompts from calibration_coding suite, max_tokens=192, median tok/s. tok_s_e2e = generated_tokens/total_elapsed (wall-clock, includes prefill). tok_s_decode = decode-only. multiplier_vs_ar uses e2e. The prior baked numbers (acceptance 1.0/0.98/0.94, tok/s 63) were on Apple M5 Max 128GB and are not reproducible on M5 Pro 64GB due to lower memory bandwidth; preserved below as historical_m5max. Acceptance is stable across runs (greedy draft gives 0.87/0.74/0.64, recommended 0.7 draft gives 0.90/0.76/0.65 -- within noise).",
+    "historical_m5max": {
+      "acceptance_by_depth": [
+        1.0,
+        0.9795918367346939,
+        0.9387755102040817
+      ],
       "accepted_by_depth": [
         49,
+        48,
+        46
       ],
+      "correction_tokens": 3,
+      "depth": 3,
+      "draft_lm_head": "3-bit affine group64 draft-only head",
       "draft_sampler": {
+        "temperature": 0.7,
+        "top_k": 20,
+        "top_p": 0.95
       },
       "drafted_by_depth": [
+        49,
+        49,
+        49
       ],
+      "enable_thinking": false,
+      "greedy_diagnostic": {
+        "accepted_by_depth": [
+          49,
+          49,
+          43
+        ],
+        "correction_tokens": 0,
+        "draft_sampler": {
+          "temperature": 0.0,
+          "top_k": 0,
+          "top_p": 1.0
+        },
+        "drafted_by_depth": [
+          51,
+          51,
+          51
+        ],
+        "target_sampler": {
+          "temperature": 0.0,
+          "top_k": 0,
+          "top_p": 1.0
+        },
+        "tok_s": 60.108346853627076,
+        "verify_calls": 51
       },
+      "older_3bit_tok_s": [
+        60.03822277200243,
+        60.06063543439426
+      ],
+      "previous_best_tok_s": [
+        61.52744855840259,
+        60.89966247416089
+      ],
+      "profile": "sustained",
+      "timestamp": "2026-05-03T23:07:00+0100",
+      "tok_s": [
+        63.05651084048216,
+        62.885564723941826
+      ],
+      "verify_calls": 49
+    }
   },
   "verified_on": {
+    "hardware": "Apple M5 Pro",
     "machine_arch": "arm64",
+    "macos": "26.5.2",
+    "model": "Qwen3.6-27B-MTPLX-Optimized-Speed-Vision",
+    "timestamp": "2026-07-18T00:00:00+00:00"
   }
 }