Kwaipilot/KAT-Coder-V2.5-Dev

Kwaipilot/KAT-Coder-V2.5-Dev is listed on Hugging Face as a text-generation model. This page records the public repository snapshot 7be56fe773e72b6f5ca93c1ae45d828ddb893922; verify the listed revision, licence, and runtime support before using it.

Updated

Computraq is preparing for launch. These guides explain readiness and the proposed framework; they do not establish a live service, integration or partnership.

Identity and access

The publisher record identifies this repository as Kwaipilot/KAT-Coder-V2.5-Dev. Its current public snapshot is 7be56fe773e72b6f5ca93c1ae45d828ddb893922. Access is public repository; review the applicable licence and publisher terms; the recorded licence label is apache-2.0. A public repository or gated download does not establish permission for every use.

Configuration facts

The pinned configuration reports model type qwen3_5_moe and architecture Qwen3_5MoeForConditionalGeneration. The model hub tags it for text-generation; these labels describe repository metadata and do not establish benchmark performance.

  • text_config.head_dim: 256
  • text_config.hidden_size: 2048
  • text_config.max_position_embeddings: 262144
  • text_config.num_attention_heads: 16
  • text_config.num_experts: 256
  • text_config.num_experts_per_tok: 8
  • text_config.num_hidden_layers: 40
  • text_config.num_key_value_heads: 2
  • text_config.rms_norm_eps: 1e-06
  • text_config.rope_parameters.rope_theta: 10000000
  • text_config.vocab_size: 248320
  • vision_config.hidden_size: 1152
  • vision_config.intermediate_size: 4304

Model-specific qualification

This configuration or publisher metadata marks the model as a mixture of experts. Confirm whether the serving runtime supports its routing layers, and distinguish total parameters from parameters active per token; no active count is inferred here unless the publisher explicitly states one.

The repository metadata identifies a multimodal model (image). Check the processor, image/audio/video pathways, input limits, and serving-runtime support; text-only token throughput is not a complete measure of this workload.

Weight-memory planning

This guide does not retain an independently verified logical parameter count, so it omits a numeric weight-memory estimate. If a count is verified for this exact repository revision, use parameter count × bytes per parameter for a weights-only floor; then account separately for KV cache, context, activations, runtime, and concurrency.

Revision and reproducibility

The metadata and configuration are pinned to revision 7be56fe773e72b6f5ca93c1ae45d828ddb893922. Pin this commit when reproducing an evaluation. Repository branches can change, and this source check does not validate the model weights themselves.

Published record

  • Task: text-generation
  • Model type: qwen3_5_moe
  • Architecture: Qwen3_5MoeForConditionalGeneration
  • Licence label: apache-2.0
  • Access: public repository; review the applicable licence and publisher terms
  • Revision: 7be56fe773e72b6f5ca93c1ae45d828ddb893922

Sources and verification