Part of the ADI (Advanced Data Intelligence) model line — ADI Llama series.
An uncensored, fully local model that reasons and answers like a frontier
teacher. Built by distilling glm-5.2 general-knowledge responses into an
abliterated Llama-3.1-8B student with a light 4-bit QLoRA fine-tune, then
merged, converted, and quantized to GGUF. The base is an abliterated
(refusal-suppressed) Meta-Llama-3.1-8B-Instruct, and the distillation was designed
to add the teacher's answer quality without restoring refusal behavior. The
student base retains native tool calling and a long context window.
ollama run hf.co/AdvancedDataIntelligence/adi-llama-3.1-8b-ablit-glm5.2-GGUF:Q4_K_M
Or download the .gguf and point any llama.cpp-based runtime at it.
What this model is
This is a knowledge distillation: a strong teacher (glm-5.2) generated
high-quality answers across a clean general-knowledge prompt set, and the
abliterated Llama-3.1-8B student was fine-tuned to imitate them. The result reasons
and responds more like its teacher on general topics while keeping the base's
uncensored character.
What distillation does — and doesn't do. It transfers the teacher's
reasoning style and answer quality, not net-new facts. For raw factual recall,
retrieval-augmented generation (RAG) is the right tool, not fine-tuning. What you
get here is an 8B that structures and explains like a larger model on topics it
already partly knows — without the refusal behavior of an aligned model.
Uncensored behavior — please read
This model is built on an abliterated base: the refusal direction has been
suppressed, so it will attempt most requests rather than declining them. The
fine-tune was intentionally kept light (2 epochs, benign-only data) to avoid
re-introducing refusals, and post-training spot checks confirmed the model still
answers helpfully without added hedging.
You are responsible for using it lawfully and ethically. It has weaker built-in
safety guardrails than stock Meta-Llama-3.1-8B-Instruct; apply your own filtering
and oversight for any production or public-facing deployment.
Training
Metric
Value
Training pairs
2,000 (deterministic subset of a 4,982-pair clean set)
Epochs
2 (kept light to preserve abliteration)
Steps
500
Final train loss
1.1143
LoRA rank / alpha
16 / 16
Trainable params
41.9M
Precision
4-bit QLoRA (nf4)
Peak VRAM
9.66 GB
Hardware
single RTX 5060 Ti (16 GB)
Training time
1.73 h (~12 s/step)
The seed prompts were drawn from the human-written
Databricks Dolly-15k
dataset (filtered to remove items requiring an attached context passage, then
deduplicated). The teacher was queried with thinking disabled so the student
learns clean final answers rather than chain-of-thought.
Notes for re-builders
Distilling onto an abliterated base is a balancing act. Any SFT can nudge an
abliterated model back toward refusals. Two choices kept the uncensored behavior
intact: benign-only training data (the GLM-5.2 set has zero refusals to
re-learn) and a light touch (LoRA rank 16, 2 epochs). Spot-check refusals
before/after.
Llama 3.1 is a standard LlamaForCausalLM — no arch surprises. 4-bit QLoRA
via Unsloth with gradient checkpointing ("unsloth" mode), max_seq_length 2048,
per-device batch 2 × grad-accum 4, adamw_8bit, LoRA targeting all attention + MLP
projections. Peak VRAM 9.66 GB on a 16 GB card.
GGUF conversion via streaming LoRA merge → f16 GGUF → Q4_K_M quantize with
llama.cpp (convert_hf_to_gguf.py). Use the standard Llama-3.1 chat template
(<|start_header_id|> / <|eot_id|>) in the Modelfile.
Intended use
General-purpose local assistant for users who want a capable, private,
offline-capable model with minimal refusal behavior: explanations, reasoning,
creative writing, and tool-calling workflows. Not intended as a source of
authoritative facts without retrieval, and not a substitute for your own safety
review.
License
Llama 3.1 Community License, inherited from the Llama-3.1-8B lineage via the
abliterated
base model.
Review the Llama 3.1 license and
Acceptable Use Policy — note the
attribution requirement ("Built with Llama") and use restrictions. Distilled
training data was generated using glm-5.2; users should review the teacher model's
terms for their own use case.
Built at theLAB — Learning. Algorithms. Breakthroughs.
