[Desktop] weights failed to load: Qwen3.6-35B-A3B NVFP4 #812

Before you start

• I have checked the supported models and this checkpoint is not listed there. A listed checkpoint that fails is a bug; use a Bug report instead.
• I have read the Roadmap and this model is not already planned there.
• I have read the FAQ.
• I have searched existing issues and found no duplicate.
• This is not a GGUF checkpoint. GGUF support is on the Roadmap; please do not open an issue for it.

Hugging Face link

[15:46:07] cmd/INFO detecting CPU/PCIe bandwidth to pick the MoE backend (ft bench bw)… [15:47:42] cmd/INFO ft serve --model D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 --port 1919 --moe-backend auto --max-running-requests 4 --memory-ratio 0.85 --host 127.0.0.1 --cors-origins tauri://localhost,http://tauri.localhost,http://localhost:1420 --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:47:42] stdout/INFO serve started (pid=2840 model=D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 port=1919) [15:47:52] stdout/WARN [2026-10-23|15:47:51] WARNING --moe-backend is deprecated; use --moe-strategy [15:47:52] stdout/INFO [2026-10-23|15:47:52] INFO Parsed arguments: [15:47:52] stdout/INFO ServerArgs(model_path='D:\FreeModel\Qwen3.6-35B-A3B-NVFP4', tp_info=DistributedInfo(rank=0, size=1), dtype=torch.bfloat16, hf_overrides={}, max_running_req=4, attention_backend='auto', moe_strategy='auto', quant_backend=None, ple_backend='disk', swa_decoder_replay='bounded', expert_load='auto', moe_cache_size=0, moe_cache_rate=None, moe_cache_auto=False, moe_cache_mem=None, moe_cache_scope='unified', moe_cache_split='equal', kv_reserve_tokens=8192, moe_cache_policy='lru', moe_prefill_overlap=True, moe_prefill_hit_d2d=False, moe_collect_stats=False, moe_cpu_threads=0, moe_cpu_layers=None, moe_hybrid_max_fetch=-1, moe_locked_fetch=-1.0, moe_locked_max_fetch=8, cuda_graph_bs=None, cuda_graph_max_bs=None, page_size=1, memory_ratio=0.85, linear_state_cache_ratio=2.0, swa_full_tokens_ratio=0.2, swa_num_pages_override=None, distributed_timeout=60.0, use_dummy_weight=False, use_pynccl=True, max_seq_len_override=None, num_page_override=None, num_token_override=None, mm=MultimodalConfig(disabled_encoders=frozenset(), embed_cache_device='cpu', encoder_weights='host', image_min_tokens=None, image_max_tokens=None, processor_kwargs={}), max_extend_tokens=8192, cache_type='radix', offline_mode=False, decode_log_interval=40, special_token_ckpt=False, _unique_suffix='.pid=4944', _ipc_base_port=35282, server_host='127.0.0.1', server_port=1919, num_tokenizer=0, silent_output=False, shell_mode=False, served_model_name='Qwen3.6-35B-A3B-NVFP4', tool_call_parser='qwen3_coder', reasoning_parser='qwen3', sampling_defaults='model', max_output_tokens=None, enable_cache_report=False, anthropic_inline_system='auto', allowed_media_domains='', allowed_local_media_path='', cors_origins='tauri://localhost,http://tauri.localhost,http://localhost:1420', gpu=('GPU-a507b142-ed89-28f6-a004-43e87d615dae',), gpu_assigned=None) [15:47:52] stdout/INFO [2026-10-23|15:47:52|initializer] INFO --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae -> GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:47:52] stdout/INFO [2026-10-23|15:47:52|FrontendAPI] INFO Default sampling config (source=model): temperature=1.0, top_k=20, top_p=0.95 [15:47:52] stdout/INFO INFO: Started server process [4944] [15:47:52] stdout/INFO INFO: Waiting for application startup. [15:47:52] stdout/INFO INFO: Application startup complete. [15:47:52] stdout/INFO INFO: Uvicorn running on http://127.0.0.1:1919 (Press CTRL+C to quit) [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO Auto-selected attention backend: triton [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO Auto-selected MoE strategy: offload [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO No MoE cache sizing flag given; defaulting to --moe-cache-auto for auto-selected strategy 'offload' [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO expert banks exceed the pin budget: --moe-cpu-layers defaults to 'auto' on native Windows [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO Resolved config: moe_strategy='offload', attention_backend='triton', cache_type='hybrid_radix', page_size=1 [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO Free memory before loading model: 6.93 GiB [15:48:01] stdout/INFO [2026-10-23|15:48:01|core|rank=0] INFO kernel triton selected; skipped b12x: the weight declares no static activation scale (W4A16 checkpoint); cublas: the weight declares no static activation scale (W4A16 checkpoint) (×81) [15:48:01] stdout/INFO Process freetoken-TP0-scheduler: [15:48:01] stdout/ERROR [2026-10-23|15:48:01|FrontendAPI] ERROR Backend supervisor: WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:48:01] stdout/INFO Traceback (most recent call last): [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 386, in freetoken.engine.engine._weight_load_context [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 643, in freetoken.engine.engine.Engine._load_weights [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 655, in freetoken.engine.engine.Engine._load_weight_state_dict [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 403, in freetoken.engine.engine._materialize_loaded_weight_state_dict [15:48:01] stdout/INFO File "python/freetoken/models/weight.py", line 246, in load_weight [15:48:01] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 523, in iter_ftw_weights [15:48:01] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 352, in freetoken.checkpoint.ftw.FTWReader.init [15:48:01] stdout/INFO ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:48:01] stdout/INFO [15:48:01] stdout/INFO The above exception was the direct cause of the following exception: [15:48:01] stdout/INFO [15:48:01] stdout/INFO Traceback (most recent call last): [15:48:01] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 314, in _bootstrap [15:48:01] stdout/INFO self.run() [15:48:01] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 108, in run [15:48:01] stdout/INFO self._target(*self._args, **self._kwargs) [15:48:01] stdout/INFO File "C:\Users\WGB\AppData\Local\FreeToken\venv\Lib\site-packages\freetoken\server\launch.py", line 81, in _run_scheduler [15:48:01] stdout/INFO scheduler = Scheduler(args) [15:48:01] stdout/INFO ^^^^^^^^^^^^^^^ [15:48:01] stdout/INFO File "python/freetoken/scheduler/scheduler.py", line 68, in freetoken.scheduler.scheduler.Scheduler.init [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 459, in freetoken.engine.engine.Engine.init [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 642, in freetoken.engine.engine.Engine._load_weights [15:48:01] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\contextlib.py", line 158, in exit [15:48:01] stdout/INFO self.gen.throw(value) [15:48:01] stdout/INFO File "python/freetoken/engine/engine.py", line 393, in _weight_load_context [15:48:01] stdout/INFO freetoken.engine.engine.WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:48:11] health/ERROR WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint (×5) [15:48:11] stdout/ERROR [2026-10-23|15:48:11|FrontendAPI] ERROR Backend worker is gone and cannot be restarted; stopping the API server [15:48:11] stdout/INFO serve exited with code 15 (pid=2840) [15:48:40] cmd/INFO ft serve --model D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 --port 1919 --moe-backend auto --max-running-requests 4 --memory-ratio 0.85 --host 127.0.0.1 --cors-origins tauri://localhost,http://tauri.localhost,http://localhost:1420 --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:48:40] stdout/INFO serve started (pid=36432 model=D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 port=1919) [15:48:47] stdout/WARN [2026-10-23|15:48:47] WARNING --moe-backend is deprecated; use --moe-strategy [15:48:47] stdout/INFO [2026-10-23|15:48:47] INFO Parsed arguments: [15:48:47] stdout/INFO ServerArgs(model_path='D:\FreeModel\Qwen3.6-35B-A3B-NVFP4', tp_info=DistributedInfo(rank=0, size=1), dtype=torch.bfloat16, hf_overrides={}, max_running_req=4, attention_backend='auto', moe_strategy='auto', quant_backend=None, ple_backend='disk', swa_decoder_replay='bounded', expert_load='auto', moe_cache_size=0, moe_cache_rate=None, moe_cache_auto=False, moe_cache_mem=None, moe_cache_scope='unified', moe_cache_split='equal', kv_reserve_tokens=8192, moe_cache_policy='lru', moe_prefill_overlap=True, moe_prefill_hit_d2d=False, moe_collect_stats=False, moe_cpu_threads=0, moe_cpu_layers=None, moe_hybrid_max_fetch=-1, moe_locked_fetch=-1.0, moe_locked_max_fetch=8, cuda_graph_bs=None, cuda_graph_max_bs=None, page_size=1, memory_ratio=0.85, linear_state_cache_ratio=2.0, swa_full_tokens_ratio=0.2, swa_num_pages_override=None, distributed_timeout=60.0, use_dummy_weight=False, use_pynccl=True, max_seq_len_override=None, num_page_override=None, num_token_override=None, mm=MultimodalConfig(disabled_encoders=frozenset(), embed_cache_device='cpu', encoder_weights='host', image_min_tokens=None, image_max_tokens=None, processor_kwargs={}), max_extend_tokens=8192, cache_type='radix', offline_mode=False, decode_log_interval=40, special_token_ckpt=False, _unique_suffix='.pid=23456', _ipc_base_port=30250, server_host='127.0.0.1', server_port=1919, num_tokenizer=0, silent_output=False, shell_mode=False, served_model_name='Qwen3.6-35B-A3B-NVFP4', tool_call_parser='qwen3_coder', reasoning_parser='qwen3', sampling_defaults='model', max_output_tokens=None, enable_cache_report=False, anthropic_inline_system='auto', allowed_media_domains='', allowed_local_media_path='', cors_origins='tauri://localhost,http://tauri.localhost,http://localhost:1420', gpu=('GPU-a507b142-ed89-28f6-a004-43e87d615dae',), gpu_assigned=None) [15:48:47] stdout/INFO [2026-10-23|15:48:47|initializer] INFO --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae -> GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:48:47] stdout/INFO [2026-10-23|15:48:47|FrontendAPI] INFO Default sampling config (source=model): temperature=1.0, top_k=20, top_p=0.95 [15:48:47] stdout/INFO INFO: Started server process [23456] [15:48:47] stdout/INFO INFO: Waiting for application startup. [15:48:47] stdout/INFO INFO: Application startup complete. [15:48:47] stdout/INFO INFO: Uvicorn running on http://127.0.0.1:1919 (Press CTRL+C to quit) [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO Auto-selected attention backend: triton [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO Auto-selected MoE strategy: offload [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO No MoE cache sizing flag given; defaulting to --moe-cache-auto for auto-selected strategy 'offload' [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO expert banks exceed the pin budget: --moe-cpu-layers defaults to 'auto' on native Windows [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO Resolved config: moe_strategy='offload', attention_backend='triton', cache_type='hybrid_radix', page_size=1 [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO Free memory before loading model: 6.93 GiB [15:48:55] stdout/INFO [2026-10-23|15:48:55|core|rank=0] INFO kernel triton selected; skipped b12x: the weight declares no static activation scale (W4A16 checkpoint); cublas: the weight declares no static activation scale (W4A16 checkpoint) (×81) [15:48:55] stdout/INFO Process freetoken-TP0-scheduler: [15:48:55] stdout/ERROR [2026-10-23|15:48:55|FrontendAPI] ERROR Backend supervisor: WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:48:55] stdout/INFO Traceback (most recent call last): [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 386, in freetoken.engine.engine._weight_load_context [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 643, in freetoken.engine.engine.Engine._load_weights [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 655, in freetoken.engine.engine.Engine._load_weight_state_dict [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 403, in freetoken.engine.engine._materialize_loaded_weight_state_dict [15:48:55] stdout/INFO File "python/freetoken/models/weight.py", line 246, in load_weight [15:48:55] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 523, in iter_ftw_weights [15:48:55] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 352, in freetoken.checkpoint.ftw.FTWReader.init [15:48:55] stdout/INFO ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:48:55] stdout/INFO [15:48:55] stdout/INFO The above exception was the direct cause of the following exception: [15:48:55] stdout/INFO [15:48:55] stdout/INFO Traceback (most recent call last): [15:48:55] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 314, in _bootstrap [15:48:55] stdout/INFO self.run() [15:48:55] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 108, in run [15:48:55] stdout/INFO self._target(*self._args, **self._kwargs) [15:48:55] stdout/INFO File "C:\Users\WGB\AppData\Local\FreeToken\venv\Lib\site-packages\freetoken\server\launch.py", line 81, in _run_scheduler [15:48:55] stdout/INFO scheduler = Scheduler(args) [15:48:55] stdout/INFO ^^^^^^^^^^^^^^^ [15:48:55] stdout/INFO File "python/freetoken/scheduler/scheduler.py", line 68, in freetoken.scheduler.scheduler.Scheduler.init [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 459, in freetoken.engine.engine.Engine.init [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 642, in freetoken.engine.engine.Engine._load_weights [15:48:55] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\contextlib.py", line 158, in exit [15:48:55] stdout/INFO self.gen.throw(value) [15:48:55] stdout/INFO File "python/freetoken/engine/engine.py", line 393, in _weight_load_context [15:48:55] stdout/INFO freetoken.engine.engine.WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:49:05] health/ERROR WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint (×5) [15:49:05] stdout/ERROR [2026-10-23|15:49:05|FrontendAPI] ERROR Backend worker is gone and cannot be restarted; stopping the API server [15:49:05] stdout/INFO serve exited with code 15 (pid=36432) [15:49:51] cmd/INFO ft serve --model D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 --port 1919 --moe-backend auto --max-running-requests 4 --memory-ratio 0.85 --host 127.0.0.1 --cors-origins tauri://localhost,http://tauri.localhost,http://localhost:1420 --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:49:51] stdout/INFO serve started (pid=34048 model=D:\FreeModel\Qwen3.6-35B-A3B-NVFP4 port=1919) [15:49:58] stdout/WARN [2026-10-23|15:49:58] WARNING --moe-backend is deprecated; use --moe-strategy [15:49:59] stdout/INFO [2026-10-23|15:49:58] INFO Parsed arguments: [15:49:59] stdout/INFO ServerArgs(model_path='D:\FreeModel\Qwen3.6-35B-A3B-NVFP4', tp_info=DistributedInfo(rank=0, size=1), dtype=torch.bfloat16, hf_overrides={}, max_running_req=4, attention_backend='auto', moe_strategy='auto', quant_backend=None, ple_backend='disk', swa_decoder_replay='bounded', expert_load='auto', moe_cache_size=0, moe_cache_rate=None, moe_cache_auto=False, moe_cache_mem=None, moe_cache_scope='unified', moe_cache_split='equal', kv_reserve_tokens=8192, moe_cache_policy='lru', moe_prefill_overlap=True, moe_prefill_hit_d2d=False, moe_collect_stats=False, moe_cpu_threads=0, moe_cpu_layers=None, moe_hybrid_max_fetch=-1, moe_locked_fetch=-1.0, moe_locked_max_fetch=8, cuda_graph_bs=None, cuda_graph_max_bs=None, page_size=1, memory_ratio=0.85, linear_state_cache_ratio=2.0, swa_full_tokens_ratio=0.2, swa_num_pages_override=None, distributed_timeout=60.0, use_dummy_weight=False, use_pynccl=True, max_seq_len_override=None, num_page_override=None, num_token_override=None, mm=MultimodalConfig(disabled_encoders=frozenset(), embed_cache_device='cpu', encoder_weights='host', image_min_tokens=None, image_max_tokens=None, processor_kwargs={}), max_extend_tokens=8192, cache_type='radix', offline_mode=False, decode_log_interval=40, special_token_ckpt=False, _unique_suffix='.pid=36832', _ipc_base_port=45441, server_host='127.0.0.1', server_port=1919, num_tokenizer=0, silent_output=False, shell_mode=False, served_model_name='Qwen3.6-35B-A3B-NVFP4', tool_call_parser='qwen3_coder', reasoning_parser='qwen3', sampling_defaults='model', max_output_tokens=None, enable_cache_report=False, anthropic_inline_system='auto', allowed_media_domains='', allowed_local_media_path='', cors_origins='tauri://localhost,http://tauri.localhost,http://localhost:1420', gpu=('GPU-a507b142-ed89-28f6-a004-43e87d615dae',), gpu_assigned=None) [15:49:59] stdout/INFO [2026-10-23|15:49:59|initializer] INFO --gpu GPU-a507b142-ed89-28f6-a004-43e87d615dae -> GPU-a507b142-ed89-28f6-a004-43e87d615dae [15:49:59] stdout/INFO [2026-10-23|15:49:59|FrontendAPI] INFO Default sampling config (source=model): temperature=1.0, top_k=20, top_p=0.95 [15:49:59] stdout/INFO INFO: Started server process [36832] [15:49:59] stdout/INFO INFO: Waiting for application startup. [15:49:59] stdout/INFO INFO: Application startup complete. [15:49:59] stdout/INFO INFO: Uvicorn running on http://127.0.0.1:1919 (Press CTRL+C to quit) [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO Auto-selected attention backend: triton [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO Auto-selected MoE strategy: offload [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO No MoE cache sizing flag given; defaulting to --moe-cache-auto for auto-selected strategy 'offload' [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO expert banks exceed the pin budget: --moe-cpu-layers defaults to 'auto' on native Windows [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO Resolved config: moe_strategy='offload', attention_backend='triton', cache_type='hybrid_radix', page_size=1 [15:50:06] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO Free memory before loading model: 6.93 GiB [15:50:07] stdout/INFO [2026-10-23|15:50:06|core|rank=0] INFO kernel triton selected; skipped b12x: the weight declares no static activation scale (W4A16 checkpoint); cublas: the weight declares no static activation scale (W4A16 checkpoint) (×81) [15:50:07] stdout/INFO Process freetoken-TP0-scheduler: [15:50:07] stdout/ERROR [2026-10-23|15:50:06|FrontendAPI] ERROR Backend supervisor: WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:50:07] stdout/INFO Traceback (most recent call last): [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 386, in freetoken.engine.engine._weight_load_context [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 643, in freetoken.engine.engine.Engine._load_weights [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 655, in freetoken.engine.engine.Engine._load_weight_state_dict [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 403, in freetoken.engine.engine._materialize_loaded_weight_state_dict [15:50:07] stdout/INFO File "python/freetoken/models/weight.py", line 246, in load_weight [15:50:07] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 523, in iter_ftw_weights [15:50:07] stdout/INFO File "python/freetoken/checkpoint/ftw.py", line 352, in freetoken.checkpoint.ftw.FTWReader.init [15:50:07] stdout/INFO ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:50:07] stdout/INFO [15:50:07] stdout/INFO The above exception was the direct cause of the following exception: [15:50:07] stdout/INFO [15:50:07] stdout/INFO Traceback (most recent call last): [15:50:07] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 314, in _bootstrap [15:50:07] stdout/INFO self.run() [15:50:07] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\multiprocessing\process.py", line 108, in run [15:50:07] stdout/INFO self._target(*self._args, **self._kwargs) [15:50:07] stdout/INFO File "C:\Users\WGB\AppData\Local\FreeToken\venv\Lib\site-packages\freetoken\server\launch.py", line 81, in _run_scheduler [15:50:07] stdout/INFO scheduler = Scheduler(args) [15:50:07] stdout/INFO ^^^^^^^^^^^^^^^ [15:50:07] stdout/INFO File "python/freetoken/scheduler/scheduler.py", line 68, in freetoken.scheduler.scheduler.Scheduler.init [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 459, in freetoken.engine.engine.Engine.init [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 642, in freetoken.engine.engine.Engine._load_weights [15:50:07] stdout/INFO File "C:\Users\WGB\AppData\Roaming\uv\python\cpython-3.12-windows-x86_64-none\Lib\contextlib.py", line 158, in exit [15:50:07] stdout/INFO self.gen.throw(value) [15:50:07] stdout/INFO File "python/freetoken/engine/engine.py", line 393, in _weight_load_context [15:50:07] stdout/INFO freetoken.engine.engine.WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint [15:50:15] health/ERROR WeightLoadError: ValueError: 'D:\FreeModel\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp32; reconvert it from the source checkpoint with ft checkpoint (×5) [15:50:17] stdout/ERROR [2026-10-23|15:50:16|FrontendAPI] ERROR Backend worker is gone and cannot be restarted; stopping the API server [15:50:17] stdout/INFO serve exited with code 15 (pid=34048)

Is the model architecture already supported

Yes, but this checkpoint or quantization does not load

Is the quantization already supported

Not sure

What happens when you load it

>>>>>>>> PASTE THE FULL SERVER LOG HERE <<<<<<<<
(copied with the "Copy server log" button in the failure dialog, or Logs -> Server status -> Copy)

WeightLoadError: ValueError: 'D:\\FreeModel\\Qwen3.6-35B-A3B-NVFP4' stores NVFP4 global scales as fp16 (e.g. 'model.layers.0.mlp.shared_expert.down_proj.weight_global') but FreeToken reads them as fp3

Anything else

Desktop 0.2.0-beta.23; engine unknown; NVIDIA GeForce RTX 4060 Laptop GPU, 8.00 GiB, driver 617.42; 13th Gen Intel Core i7-13700H, 31.7 GiB