Ablagent-35B-A3B

Blackfrost Pro Series

Ablagent-35B-A3B · Professional Model Engineering Toolkit

Private toolkit preview. Ablagent-35B-A3B is a task-specific workflow model packaged with a bounded standalone harness. It is built for professional model-engineering work—not as a general chat assistant, broad coding model, or autonomous weight-editing system.

The Pro Series

Blackfrost Pro Series is a line of specialized models packaged as professional tools and workflows. Each release is designed around a defined operational role instead of open-ended chat.

The series format combines:

1. a domain-tuned model;
2. a deliberate operating boundary;
3. a harness or integration surface;
4. deployment guidance;
5. a validation and provenance record.

This structure is intended to extend across model engineering, marketing operations, logistics, and other bounded professional workflows while preserving a recognizable Blackfrost product and documentation standard.

Product brief

Ablagent-35B-A3B is the Blackfrost Pro Series toolkit for Abliteration and Domain Weight Modification workflows. It helps an operator inspect artifacts, reason across run evidence, compare candidates, prepare structured plans, diagnose incomplete experiments, and produce clear handoff records.

The model is delivered with Blackfrost Ablagent Standalone Harness 1.0.0, which turns the checkpoint into a bounded, local-first workflow assistant. The current harness release is a read-only evaluation preview: it can inspect an authorized workspace through a small tool surface, but it cannot execute shell commands, modify weights, deploy services, publish artifacts, or write into the workspace.

Toolkit profile

Professional workflow

Ablagent is tuned to assist with the work surrounding a controlled weight-modification program:

• Provenance and intake — checkpoint archaeology, lineage, revision, and source-fingerprint review;
• Capture preparation — paired-corpus hygiene, role leakage, template leakage, and prompt provenance checks;
• Direction and edit review — capture interpretation, edit-surface enumeration, target-set review, and MTP preservation discipline;
• Candidate evaluation — geometry, distribution drift, behavioral evidence, and HITL comparisons across alpha/pass candidates;
• Run recovery — reconcile manifests, receipts, controls, and filesystem evidence without silently inventing missing state;
• Release operations — runtime diagnosis, structured tool-use checks, packaging review, and evidence-focused handoff records.

The intended output is not merely an answer. It is a more legible, auditable workflow for the human operator making the decision.

Operating boundary

• Ablagent does not receive access to files, weights, GPUs, endpoints, or external systems unless an operator explicitly connects them.
• The model may draft plans, diagnostics, scripts, and reports, but those outputs require review.
• Structural correctness, behavior, distribution drift, and operator acceptance remain separate gates.
• Missing measurements are recorded as missing; they are not treated as zero or as passing.
• A served candidate is not automatically an accepted or promoted checkpoint.

What this product is not

• not a general-purpose chat assistant;
• not positioned as a replacement for larger open-domain models;
• not an autonomous DWM engine;
• not a substitute for the actual checkpoint, tensor inventory, manifests, or measurements;
• not a claim that every backend, context length, tool schema, or task has been validated.

Why 35B-A3B

The 35B-A3B MoE class gives the toolkit more room for dense artifact comparison and multi-step workflow reasoning than the compact 9B companion while keeping active inference substantially smaller than a dense 35B deployment. The checkpoint was selected for a dedicated professional workflow—not to compete with the largest general-purpose assistants on broad benchmark coverage.

Technical profile

The inherited architecture contains vision components. This specialization and the recorded release validation are text-focused; multimodal quality was not re-benchmarked.

Lineage

1. Qwen/Qwen3.6-35B-A3B, pinned at revision 995ad96eacd98c81ed38be0c5b274b04031597b0
2. Ablagent specialist tuning, using 250 curated reasoning traces for two passes
3. Blackfrost DWM post-processing, accepted by the operator after side-by-side HITL
4. Ablagent-35B-A3B, the complete merged and validated BF16 checkpoint in this repository

The specialist identity is carried by the tuned checkpoint and its packaged chat template:

You are Ablagent, an AI assistant for Abliteration and DWM workflows, built by Blackfrost_AI/SI.

This repository does not include the private corpus, raw traces, LoRA adapter, optimizer state, training scripts, training recipe, model-weight DWM implementation recipe, direction bank, direction vectors, capture prompts, or internal infrastructure records.

Release validation

The exact promoted checkpoint passed the recorded Blackfrost structural, geometry, held-out distribution, runtime, packaging, and operator-acceptance gates.

These figures establish release identity and bounded validation for this specific run. They are not a claim of universal capability, correctness, or suitability for every deployment.

Included toolkit

Model checkpoint

• complete merged BF16 weights;
• Ablagent identity carried by the tuned checkpoint and packaged chat template;
• separate reasoning and final-answer fields on the validated serving stack;
• OpenAI-compatible Chat Completions deployment target;
• no adapter required at inference time.

Standalone harness

HARNESS/ contains Blackfrost Ablagent Standalone Harness 1.0.0. The same harness supports both private Ablagent releases:

• Blackfrost-AI/Ablagent-9B
• Blackfrost-AI/Ablagent-35B-A3B

It is a local-first, read-only evaluation preview with an OpenAI-compatible client, workspace boundary, staged-plan driver, receipt checks, transcript controls, and the read_file, grep, and glob tools. It exposes no shell, write, deployment, publication, or weight-editing tools.

That restriction is part of the product definition, not a missing feature in the model card. The harness presents Ablagent as a focused workflow assistant rather than an open-ended chatbot. Set the harness model field to the exact identifier returned by the selected serving endpoint.

For the 35B checkpoint:

cp HARNESS/config.35b.example.json HARNESS/config.local.json
HARNESS/bin/ablagent --config HARNESS/config.local.json --probe

Then run a bounded read-only workspace task:

HARNESS/bin/ablagent \
  --config HARNESS/config.local.json \
  --workspace /path/to/workspace \
  --task "Inspect the current run state and report supported facts, conflicts, and untested claims."

Read HARNESS/docs/SECURITY.md, HARNESS/docs/READINESS.md, and HARNESS/docs/SERVER.md before broader use.

Serving contract

The validated server pairing uses:

• OpenAI-compatible Chat Completions;
• served model name Ablagent-35B-A3B;
• qwen3 reasoning parser;
• qwen3_coder tool-call parser;
• reasoning_content kept separate from final content;
• tensor parallel size 1 on the validated B300 deployment.

Preserve the packaged chat template and allow enough completion tokens for both reasoning and the final answer. Tool schemas, reasoning parser, tool-call parser, backend, and template should be validated as one serving contract.

Limitations and operator responsibilities

• Ablagent-35B-A3B is a specialist workflow model and is not presented as a replacement for larger general-purpose assistants.
• Ablagent can be wrong, incomplete, or overconfident about tensor layouts, code, and runtime behavior.
• Structural and distribution gates do not replace task-specific behavioral evaluation.
• The 262,144-token configuration is not a claim of full-length quality on every backend.
• The current harness is a read-only evaluation preview, not a production automation controller.
• Operators remain responsible for authorization, access control, data handling, validation, destructive-action review, and lawful deployment.

Repository contents

• Full accepted BF16 model weights, including the byte-preserved MTP payload
• Configuration, tokenizer, processor configuration, and Ablagent chat template
• Blackfrost Pro Series model card and banner
• Dual-model Ablagent Standalone Harness 1.0.0
• Upstream and harness license/notice files

No training recipe or model-weight DWM implementation recipe is included. The harness contains only its generic read-only workflow guidance and synthetic demonstration workspace.

License and attribution

The upstream Qwen checkpoint declares Apache License 2.0. The upstream license is included at LICENSE, with lineage recorded in UPSTREAM-LICENSE-NOTICE.md. The bundled harness is separately distributed under Apache License 2.0 and carries its own license and notice under HARNESS/.