--- license: apache-2.0 license_link: https://huggingface.co/ibm-granite/granite-4.1-3b/blob/main/LICENSE language: - en library_name: transformers pipeline_tag: text-generation base_model: AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5 tags: - granite - parable - heretic - uncensored - decensored - abliterated - reproducible - conversational - text-generation-inference - reasoning - thinking - agent - coding - tool-use - terminal - claude-fable-5 datasets: - AnkitAI/parable-corpus-v2 - Glint-Research/Fable-5-traces --- # Parable-Granite-4.1-3B-Claude-Fable-5-heretic
RACER IS OP

A decensored variant of [AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5](https://huggingface.co/AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5) (fine-tuned from [ibm-granite/granite-4.1-3b](https://huggingface.co/ibm-granite/granite-4.1-3b)), produced with [Heretic](https://github.com/p-e-w/heretic) v1.4.0 (directional ablation / "abliteration"). Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so Parable's Claude Fable 5 trace-distilled reasoning, planning, and terminal instincts are left largely intact. **Who this is for:** developers who want Parable's tiny local reasoning model — planning, code generation, terminal workflows with `` traces — without refusals. Runs in ~3 GB RAM (Q4_K_M) on laptops, old GPUs, or swap-backed boxes. Not a capability upgrade over base Parable-Granite 3B — same model, refusal guardrails removed. ## Why abliteration instead of fine-tuning Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the [Heretic repo](https://github.com/p-e-w/heretic) and the [original abliteration writeup](https://huggingface.co/blog/mlabonne/abliteration) for the mechanism. ## Runs on your gaming PC Full GGUF ladder included — pick the quant that fits your card: | Your GPU | Recommended quant | Weights | | :--- | :--- | :--- | | RTX 3060 / 4070 / 5070 (12 GB) | Q8_0 | 3.62 GB | | RTX 4060 / 3070 (8 GB) | Q6_K | 2.80 GB | | GTX 1660 Super / 2060 / 3050 laptop (6 GB) | Q5_K_M | 2.44 GB | | CPU-only / Apple Silicon | Q4_K_M | 2.10 GB, fits in system RAM | Weights only, at this model's ~3.4B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up. ## Abliteration parameters | Parameter | Value | | :-------- | :---: | | **direction_index** | 27.16 | | **attn.o_proj.max_weight** | 1.49 | | **attn.o_proj.max_weight_position** | 38.22 | | **attn.o_proj.min_weight** | 1.42 | | **attn.o_proj.min_weight_distance** | 23.26 | | **mlp.down_proj.max_weight** | 1.47 | | **mlp.down_proj.max_weight_position** | 25.37 | | **mlp.down_proj.min_weight** | 0.61 | | **mlp.down_proj.min_weight_distance** | 1.63 | ## Performance | Metric | This model | Original model ([AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5](https://huggingface.co/AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5)) | | :----- | :--------: | :---------------------------: | | **KL divergence** | 0.0111 | 0 *(by definition)* | | **Refusals** | 5/100 | 96/100 | Refusals drop from 96 to 5 out of 100 adversarial prompts at a KL cost of 0.0111 — the Parable reasoning gains (agent artifacts 0/34, Fable 5 traces) are preserved. > Made with ❤️ by **RACER IS OP** — follow for more uncensored models ## Files | File | Format | Size | |---|---|---| | `model-00001-of-00002.safetensors` + `model-00002-of-00002.safetensors` | BF16 Safetensors (sharded) | ~6.8 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-F16.gguf` | GGUF F16 | 6.81 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q2_K.gguf` | GGUF Q2_K | 1.37 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-IQ3_S.gguf` | GGUF IQ3_S | 1.57 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q3_K_S.gguf` | GGUF Q3_K_S | 1.57 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q3_K_M.gguf` | GGUF Q3_K_M | 1.73 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q3_K_L.gguf` | GGUF Q3_K_L | 1.86 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-IQ4_XS.gguf` | GGUF IQ4_XS | 1.91 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q4_K_S.gguf` | GGUF Q4_K_S | 2.00 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q4_0.gguf` | GGUF Q4_0 | 1.98 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q4_1.gguf` | GGUF Q4_1 | 2.18 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q4_K_M.gguf` | GGUF Q4_K_M | 2.10 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q5_K_S.gguf` | GGUF Q5_K_S | 2.38 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q5_K_M.gguf` | GGUF Q5_K_M | 2.44 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q6_K.gguf` | GGUF Q6_K | 2.80 GB | | `Parable-Granite-4.1-3B-Claude-Fable-5-heretic-Q8_0.gguf` | GGUF Q8_0 | 3.62 GB | | `reproduce/` | Config + eval transcripts + checksums | — | Full 15-file GGUF ladder (F16 + 14 quants) produced with [llama.cpp](https://github.com/ggml-org/llama.cpp) (`GraniteForCausalLM` architecture). The text-only GGUFs load in llama.cpp / Ollama / LM Studio / Jan directly. For transformers/vLLM use the Safetensors above. Run `llama serve -hf saidutta69/Parable-Granite-4.1-3B-Claude-Fable-5-heretic` to pull the default Q4_K_M quant. ## Quickstart ```bash # llama.cpp - defaults to Q4_K_M llama serve -hf saidutta69/Parable-Granite-4.1-3B-Claude-Fable-5-heretic:Q4_K_M # Ollama ollama run hf.co/saidutta69/Parable-Granite-4.1-3B-Claude-Fable-5-heretic:Q4_K_M ``` ```python # transformers from transformers import AutoModelForCausalLM, AutoTokenizer model_id = "saidutta69/Parable-Granite-4.1-3B-Claude-Fable-5-heretic" tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained( model_id, torch_dtype="auto", device_map="auto", ) messages = [{"role": "user", "content": "Write a Python function that retries an HTTP request with exponential backoff."}] inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device) out = model.generate(**inputs, max_new_tokens=3000, temperature=0.7, top_p=0.95, do_sample=True) print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True)) # Note: Parable emits ... reasoning blocks before the answer (Fable 5 heritage). Strip or surface them as needed. ``` Also runnable via LM Studio, Jan, vLLM, SGLang — see the "Use this model" widget above for copy-paste commands. ## Thinking mode Every answer opens with a `...` block — that's the Fable 5 trace heritage. Use llama.cpp `--jinja` to separate it automatically. Sampling: temperature 0.7, top_p 0.95, and budget 2500+ tokens for full reasoning. ## Responsible use Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it — don't put this behind an unmoderated public endpoint serving third parties. ## License Inherits the [apache-2.0](https://huggingface.co/ibm-granite/granite-4.1-3b/blob/main/LICENSE) license from the base model ([ibm-granite/granite-4.1-3b](https://huggingface.co/ibm-granite/granite-4.1-3b)) via [AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5](https://huggingface.co/AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5). ## Related - [AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5](https://huggingface.co/AnkitAI/Parable-Granite-4.1-3B-Claude-Fable-5) — the parent model - [ibm-granite/granite-4.1-3b](https://huggingface.co/ibm-granite/granite-4.1-3b) — the base model - [AnkitAI/parable-corpus-v2](https://huggingface.co/datasets/AnkitAI/parable-corpus-v2) — training corpus