[Bug] Unable to load Qwen 3.8 Flash Next (Specifically MLX) on M5 Ultra

AI truncated the log to this, more verbose log is bellow: 
"
AutoConfig probe parse failure for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' in :
ValueError: The checkpoint has model type `qwen4_exp` but Transformers does not recognize this architecture.

(same failure across .venv_t5_510, .venv_t5_530, .venv_t5_550)
Transformers tier default (5.5.0) selected — no match

Model config result: is_vision=True, max_position_embeddings=262144
Loading Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed via mlx-vlm (is_lora=False, ...)

Exception: Received 1 parameters not in model:
language_model.model.layers.1.ple.conv_weight.
"

Unedited log: 
"
01:47:08 [INFO] [backend] {"timestamp": "2026-09-29T01:47:08.943349Z", "level": "info", "event": "Detected MTP subdirectory drafter: /Users/admin/.cache/huggingface/hub/blobs/16/16e742f020c6597aeadf1fcbb3c2c7edd759a989b0c70865e94a2612014ecb5d"}
01:47:08 [INFO] [backend] {"timestamp": "2026-09-29T01:47:08.945814Z", "level": "info", "event": "Detected MTP subdirectory drafter: /Users/admin/.cache/huggingface/hub/blobs/16/16e742f020c6597aeadf1fcbb3c2c7edd759a989b0c70865e94a2612014ecb5d"}
01:50:24 [INFO] [backend] {"timestamp": "2026-09-29T01:50:24.251486Z", "level": "info", "event": "Found 0 trained models in /Users/admin/.unsloth/studio/outputs"}
01:50:24 [INFO] [backend] {"timestamp": "2026-09-29T01:50:24.260445Z", "level": "info", "event": "Found 0 exported models in /Users/admin/.unsloth/studio/exports"}
01:50:24 [INFO] [backend] {"timestamp": "2026-09-29T01:50:24.369163Z", "level": "info", "event": "Cached model scan: inspected=4 skipped_gguf=1 skipped_no_weights=0 returned=2"}
01:50:30 [INFO] [backend] {"timestamp": "2026-09-29T01:50:30.476891Z", "level": "info", "event": "Found 0 trained models in /Users/admin/.unsloth/studio/outputs"}
01:50:30 [INFO] [backend] {"timestamp": "2026-09-29T01:50:30.477233Z", "level": "info", "event": "Found 0 exported models in /Users/admin/.unsloth/studio/exports"}
01:50:33 [INFO] [backend] {"timestamp": "2026-09-29T01:50:33.542973Z", "level": "warning", "event": "llama-server did not exit on SIGTERM, sent SIGKILL"}
01:50:38 [INFO] [backend] {"timestamp": "2026-09-29T01:50:38.074145Z", "level": "info", "event": "Found 0 trained models in /Users/admin/.unsloth/studio/outputs"}
01:50:38 [INFO] [backend] {"timestamp": "2026-09-29T01:50:38.074476Z", "level": "info", "event": "Found 0 exported models in /Users/admin/.unsloth/studio/exports"}
01:50:38 [INFO] [backend] {"timestamp": "2026-09-29T01:50:38.208514Z", "level": "info", "event": "Unloaded GGUF model: unsloth/Qwen3.8-Flash-Next-GGUF"}
01:50:38 [INFO] [backend] {"timestamp": "2026-09-29T01:50:38.606839Z", "level": "info", "event": "Spawning fresh inference subprocess for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' (transformers 5.x, attempt 1/2)"}
01:50:38 [INFO] [backend] {"timestamp": "2026-09-29T01:50:38.645740Z", "level": "info", "event": "Inference subprocess started (pid=5230)"}
01:50:39 [INFO] [backend] {"timestamp": "2026-09-29T01:50:39.464552Z", "level": "info", "event": "Getting model config for: Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed"}
01:50:39 [INFO] [backend] {"timestamp": "2026-09-29T01:50:39.565184Z", "level": "info", "event": "Model config result for Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed: is_vision=True, is_embedding=False, audio_type=None, audio_type_known=True, is_lora=False, max_position_embeddings=262144"}
01:50:40 [INFO] [backend] {"timestamp": "2026-09-29T01:50:40.643382Z", "level": "info", "event": "AutoConfig probe parse failure for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' in : Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\nValueError: The checkpoint you are trying to load has model type `qwen4_exp` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.\n\nYou can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`"}
01:50:41 [INFO] [backend] {"timestamp": "2026-09-29T01:50:41.940731Z", "level": "info", "event": "AutoConfig probe parse failure for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' in /Users/admin/.unsloth/studio/.venv_t5_530: Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\nValueError: The checkpoint you are trying to load has model type `qwen4_exp` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.\n\nYou can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`"}
01:50:43 [INFO] [backend] {"timestamp": "2026-09-29T01:50:43.093276Z", "level": "info", "event": "AutoConfig probe parse failure for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' in /Users/admin/.unsloth/studio/.venv_t5_550: Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\nValueError: The checkpoint you are trying to load has model type `qwen4_exp` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.\n\nYou can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`"}
01:50:44 [INFO] [backend] {"timestamp": "2026-09-29T01:50:44.265620Z", "level": "info", "event": "AutoConfig probe parse failure for 'Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed' in /Users/admin/.unsloth/studio/.venv_t5_510: ValueError: The checkpoint you are trying to load has model type `qwen4_exp` but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.\n\nYou can update Transformers with the command `pip install --upgrade transformers`. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command `pip install git+https://github.com/huggingface/transformers.git`"}
01:50:44 [INFO] [backend] {"timestamp": "2026-09-29T01:50:44.265739Z", "level": "info", "event": "Transformers tier default selected for Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed (AutoConfig probe found no higher tier; config saved by transformers 5.x)"}
01:50:44 [INFO] [backend] {"timestamp": "2026-09-29T01:50:44.265763Z", "level": "info", "event": "Transformers tier default (5.5.0) selected for Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed (no match)"}
01:50:44 [INFO] [backend] {"timestamp": "2026-09-29T01:50:44.265799Z", "level": "info", "event": "Using default transformers (5.5.0) for Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed"}
01:50:46 [INFO] [backend] [stderr] W0928 20:50:46.005000 5230 torch/distributed/elastic/multiprocessing/redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
01:50:46 [INFO] [backend] {"timestamp": "2026-09-29T01:50:46.081095Z", "level": "info", "event": "Subprocess status: Loading model..."}
01:50:46 [INFO] [backend] Hardware detected: MLX — Apple Silicon (arm64)
01:50:46 [INFO] [backend] {"timestamp": "2026-09-29T01:50:46.621442Z", "level": "info", "event": "Loaded default model defaults from /Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/assets/configs/model_defaults/default.yaml"}
01:50:46 [INFO] [backend] {"timestamp": "2026-09-29T01:50:46.841955Z", "level": "info", "event": "MLX memory caps: memory_limit=203.27 GB, wired_limit=203.27 GB"}
01:50:46 [INFO] [backend] {"timestamp": "2026-09-29T01:50:46.842041Z", "level": "info", "event": "Loading Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed via mlx-vlm (is_lora=False, distributed=False, rank=0/1, mode=None)"}
01:50:48 [INFO] [backend] Unsloth: Loading Youssofal/Qwen3.8-Flash-Next-MTPLX-Optimized-Speed via mlx-vlm (VLM)...
01:50:48 [INFO] [backend] {"timestamp": "2026-09-29T01:50:48.072603Z", "level": "info", "event": "MLX inference subprocess ready, entering command loop"}
01:50:48 [INFO] [backend] {"timestamp": "2026-09-29T01:50:48.964255Z", "level": "info", "event": "Inference subprocess shut down"}
01:50:48 [INFO] [backend] {"timestamp": "2026-09-29T01:50:48.964441Z", "level": "error", "event": "Error loading model: Received 1 parameters not in model: \nlanguage_model.model.layers.1.ple.conv_weight.", "exception": "Traceback (most recent call last):\n  File \"/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/routes/inference.py\", line 17603, in _load_model_impl\n    success = await asyncio.to_thread(\n              ^^^^^^^^^^^^^^^^^^^^^^^^\n    ...<23 lines>...\n    )\n    ^\n  File \"/Users/admin/.local/share/uv/python/cpython-3.13.14-macos-aarch64-none/lib/python3.13/asyncio/threads.py\", line 26, in to_thread\n    return await loop.run_in_executor(None, func_call)\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/Users/admin/.local/share/uv/python/cpython-3.13.14-macos-aarch64-none/lib/python3.13/concurrent/futures/thread.py\", line 59, in run\n    result = self.fn(*self.args, **self.kwargs)\n  File \"/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/utils/gpu_memory_events.py\", line 36, in wrapper\n    return fn(*args, **kwargs)\n  File \"/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/core/inference/orchestrator.py\", line 2244, in load_model\n    raise Exception(error)\nException: Received 1 parameters not in model: \nlanguage_model.model.layers.1.ple.conv_weight."}
01:50:48 [INFO] [backend] | Traceback (most recent call last):
01:50:48 [INFO] [backend] |   File "/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/routes/inference.py", line 17603, in _load_model_impl
01:50:48 [INFO] [backend] |     success = await asyncio.to_thread(
01:50:48 [INFO] [backend] |               ^^^^^^^^^^^^^^^^^^^^^^^^
01:50:48 [INFO] [backend] |     ...<23 lines>...
01:50:48 [INFO] [backend] |     )
01:50:48 [INFO] [backend] |     ^
01:50:48 [INFO] [backend] |   File "/Users/admin/.local/share/uv/python/cpython-3.13.14-macos-aarch64-none/lib/python3.13/asyncio/threads.py", line 26, in to_thread
01:50:48 [INFO] [backend] |     return await loop.run_in_executor(None, func_call)
01:50:48 [INFO] [backend] |            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
01:50:48 [INFO] [backend] |   File "/Users/admin/.local/share/uv/python/cpython-3.13.14-macos-aarch64-none/lib/python3.13/concurrent/futures/thread.py", line 59, in run
01:50:48 [INFO] [backend] |     result = self.fn(*self.args, **self.kwargs)
01:50:48 [INFO] [backend] |   File "/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/utils/gpu_memory_events.py", line 36, in wrapper
01:50:48 [INFO] [backend] |     return fn(*args, **kwargs)
01:50:48 [INFO] [backend] |   File "/Users/admin/.unsloth/studio/unsloth_studio/lib/python3.13/site-packages/studio/backend/core/inference/orchestrator.py", line 2244, in load_model
01:50:48 [INFO] [backend] |     raise Exception(error)
01:50:48 [INFO] [backend] | Exception: Received 1 parameters not in model: 
01:50:48 [INFO] [backend] | language_model.model.layers.1.ple.conv_weight.
01:50:48 [INFO] [backend] {"timestamp": "2026-09-29T01:50:48.968002Z", "level": "info", "event": "request_completed", "method": "POST", "path": "/api/inference/load", "status_code": 500, "process_time_ms": 10689.45}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.519737Z", "level": "info", "event": "Reusing verified cached GGUF: /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/Q8_0/Qwen3.8-Flash-Next-Q8_0-00001-of-00006.gguf"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.520032Z", "level": "info", "event": "Reusing cached mmproj: /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/mmproj-F16.gguf"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.555654Z", "level": "info", "event": "Reusing cached MTP drafter: /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/MTP/mtp-Qwen3.8-Flash-Next-Q8_0.gguf"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.665057Z", "level": "info", "event": "GGUF size: 175.3 GB, mmproj: 0.8 GB, est. KV cache: 6.4 GB, MTP reserve: 11.83 GB (draft KV @ 262144 x 4 slots, ubatch 512, n_max 3), context: 262144, GPUs free: [], selected: None, --fit: on"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.665463Z", "level": "warning", "event": "A context of 262,144 tokens needs about 196 GB of unified memory, more than the 143 GB Studio budgets from the memory free right now. It fits under the GPU's wired memory limit, so it is loading anyway, but macOS may have to compress or swap other apps to make room and generation may slow down. Lower the context or free memory to avoid that."}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.667144Z", "level": "info", "event": "Using separate MTP drafter: /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/MTP/mtp-Qwen3.8-Flash-Next-Q8_0.gguf"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.667174Z", "level": "info", "event": "Spec decoding: draft-mtp (MTP-only)"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.668239Z", "level": "info", "event": "Reasoning model: {'enable_thinking': True, 'preserve_thinking': True} by default (launched as --reasoning on --chat-template-kwargs {\"preserve_thinking\": true})"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.668265Z", "level": "info", "event": "Using mmproj for vision: /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/mmproj-F16.gguf"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.668334Z", "level": "info", "event": "Capping llama-server context checkpoints at 29 per slot (this build's default 32): each one snapshots 112.2 MiB of this model's state, so the default would hold 14.0 GiB of host RAM across 4 slot(s)."}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.668444Z", "level": "info", "event": "Model Memory: applying loading policy (--load-mode mmap+mlock)"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.668474Z", "level": "info", "event": "Starting llama-server: /Users/admin/.unsloth/llama.cpp/build/bin/llama-server -m /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/Q8_0/Qwen3.8-Flash-Next-Q8_0-00001-of-00006.gguf --port 58183 --parallel 4 --flash-attn on --no-context-shift -c 262144 --video-fps 1 --alias unsloth/Qwen3.8-Flash-Next-GGUF --fit on --metrics --slot-save-path /Users/admin/.unsloth/studio/cache/llama-slots --kv-unified --fit-ctx 262144 --jinja --model-draft /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/MTP/mtp-Qwen3.8-Flash-Next-Q8_0.gguf --spec-type draft-mtp --spec-draft-n-max 3 --reasoning on --chat-template-kwargs {\"preserve_thinking\": true} --mmproj /Users/admin/.cache/huggingface/hub/models--unsloth--Qwen3.8-Flash-Next-GGUF/snapshots/38bb39ee97821de2c9009abb7e93950eec396e66/mmproj-F16.gguf --ctx-checkpoints 29 --load-mode mmap+mlock"}
01:50:49 [INFO] [backend] {"timestamp": "2026-09-29T01:50:49.669674Z", "level": "info", "event": "llama-server stdout/stderr -> /Users/admin/.unsloth/studio/logs/llama-server/llama-1790646649-port-58183-try0.log"}
01:50:58 [INFO] [backend] {"timestamp": "2026-09-29T01:50:58.895905Z", "level": "info", "event": "llama-server ready on port 58183 for model 'unsloth/Qwen3.8-Flash-Next-GGUF'"}
01:50:58 [INFO] [backend] {"timestamp": "2026-09-29T01:50:58.900315Z", "level": "info", "event": "Loaded GGUF model via llama-server: unsloth/Qwen3.8-Flash-Next-GGUF"}
01:50:58 [INFO] [backend] {"timestamp": "2026-09-29T01:50:58.924426Z", "level": "info", "event": "Found 0 trained models in /Users/admin/.unsloth/studio/outputs"}
01:50:58 [INFO] [backend] {"timestamp": "2026-09-29T01:50:58.924638Z", "level": "info", "event": "Found 0 exported models in /Users/admin/.unsloth/studio/exports"}
"