CATALOGUE RECORD / HUB-DERIVED
nvidia
Qwen3.6-35B-A3B-NVFP4
Revision 1355db6a052410cfd62085d94b58866fd0f2c3c5
INDEX INCONSISTENCY
This repository publishes two different parameter counts.
The safetensors index declares 18,683,860,336 parameters in its total, while its own per-dtype map sums to 19,528,501,104. Those two fields describe the same tensors, so one of them is wrong.
No parameter count is published for this record. Picking the more plausible of two contradictory figures would be a guess presented as a fact. The tensor byte total below is computed from the per-dtype map alone and is cross-checked against the repository’s stored bytes.
TENSOR ACCOUNTING
Where the bytes are.
Summed from the safetensors index, one row per dtype. A parameter count alone cannot produce this figure, because a checkpoint may mix widths.
| Dtype | Parameters | Bytes each | Bytes | Share |
|---|---|---|---|---|
U8 | 16,423,321,600 | 1 | 16.4 GB | 76.9% |
BF16 | 1,825,916,784 | 2 | 3.7 GB | 17.1% |
F8_E4M3 | 1,279,262,720 | 1 | 1.3 GB | 6.0% |
EVERY FIELD, WITH ITS ORIGIN
Sourced or undetermined. Never assumed.
Each value below names the exact API field it was computed from. Where the Hub does not establish a value, the reason is shown instead of a plausible default.
- Architecture
- qwen3_5_moe
config.model_type - Model classes
- Qwen3_5MoeForConditionalGeneration
config.architectures - Routed experts
- UndeterminedThe published config exposes no routed-expert count. The Hub's config summary omits fields some architectures place only in the full config.json.
- Experts per token
- UndeterminedThe published config exposes no per-token expert count.
- Shared experts
- UndeterminedThe published config declares no always-on shared experts.
- Routing sparsity
- UndeterminedRouting sparsity requires both a routed-expert and a per-token expert count.
- Total parameters
- UndeterminedThe published index is internally inconsistent: safetensors.total declares 18,683,860,336 parameters while the per-dtype map sums to 19,528,501,104. Neither figure can be treated as the parameter count.
- Checkpoint bytes
- 21,354,417,888 (21.4 GB)
safetensors.parameters - Ships below 16-bit
- Yes
safetensors.parameters - Quantisation method
- modelopt
config.quantization_config.quant_method - Trained context
- UndeterminedThe config summary omits max_position_embeddings. Trained context is a model-card claim, not a derivable fact.
- Declared licence
- apache-2.0
cardData.license - Base model
- Qwen/Qwen3.6-35B-A3B
cardData.base_model - Library
- Model Optimizer
library_name - Files in repository
- 17
siblings - Last modified
- 2026-08-29
lastModified
LICENCE POSTURE
Permissive
The repository declares a licence that is generally read as permitting commercial use. Read the licence file in the repository before relying on that.
This is a reading of a metadata field, not legal advice, and it does not account for the licences of upstream models or training data.
WEIGHT RESIDENCY FLOOR
The count below which it cannot fit.
Ceiling of checkpoint bytes over advertised accelerator memory. This is a lower bound on accelerator count for weights alone — KV cache, activations and runtime overhead all sit on top, so a real deployment needs more.