Skip to content

EthenEthenEthen

LoginPlatformTry Ethen

Open Source Model Profile · Cannae-AI

HERETICODER-2.5-7B-IT

HERETICODER-2.5-7B-IT is a 7.62B-parameter Qwen coder model from Cannae-AI. According to the model card, it is an abliterated decensored version of Qwen2.5-Coder-7B-Instruct.

Publisher
Cannae-AI
Task
text-generation
Model type
qwen2
License
apache-2.0
Library
Unknown
Publication status
Accepted · not indexed

Model overview

HERETICODER-2.5-7B-IT is published by Cannae-AI as a Qwen2 text-generation model. The captured configuration identifies Qwen2ForCausalLM with a qwen2 model type, and Safetensors metadata reports 7,615,616,512 parameters. According to the model card, it is an abliterated decensored version of Qwen/Qwen2.5-Coder-7B-Instruct built with Heretic.

Recorded capabilities

Qwen2.5-Coder base linkage

Hub tags name Qwen/Qwen2.5-Coder-7B-Instruct as the base model and fine-tune source. According to the model card, this record is its abliterated decensored version.

Heretic abliteration

According to the model card, the decensored version was produced by ablation using Heretic.

Reported refusal profile

According to the model card, the model records 1 refusal in 100 with 0.00 KL divergence, alongside per-layer Heretic direction parameters.

Apache-2.0 licensing

The captured record and the model card both state an Apache-2.0 license.

Use cases in the source record

  • Code-focused text generation building on the documented Qwen2.5-Coder-7B-Instruct base.
  • Abliteration-profile inspection using the card's reported refusal count, KL divergence, and Heretic direction parameters.

Limitations and unknowns

  • The refusal and KL-divergence figures are publisher-reported abliteration claims and were not independently verified.
  • No evaluation results, context-window value, or hardware requirements were extracted from this record.
  • No license text was captured beyond the Apache-2.0 identifier; full terms should be confirmed at the source.
  • Provider state is historical snapshot data, not independently refreshed current availability.

Source and provenance

Source: Cannae-AI/HERETICODER-2.5-7B-IT

Captured: Unknown. Processed: 2026-09-07T19:35:38.382304+00:00.

HERETICODER-2.5-7B-IT Model Description This is an abliterated decensored version of Qwen2.5-Coder-7B-Instruct using Heretic! License: apache-2.0 Abliterated by : Cannae-AI Base model : Qwen/Qwen2.5-Coder-7B-Instruct Model Refusals KL divergence HERETICODER-2.5-7B-IT 1/100 0.00 Parameters: direction_index = per layer attn.o_proj.max_weight = 1.30 attn.o_proj.max_weight_position = 16.27 attn.o_proj.min_weight = 0.74 attn.o_proj.min_weight_distance = 4.19 mlp.down_proj.max_weight = 0.98 mlp.down_proj.max_weight_position = 21.61 mlp.down_proj.min_weight = 0.54 mlp.down_proj.min_weight_distance = 2.64

F001F002F003F004F005F006F007F008F009F011F012