README.md

---
license: other
library_name: transformers
tags:
- flm
- fastflowlm
- npu
- npu2
- amd-xdna
- lemonade
base_model:
- empero-ai/Qwable-9B-Claude-Fable-5
---

# Qwable-9B-Claude-Fable-5-NPU2 (FastFlowLM / Lemonade NPU2 Quantization)

> [!IMPORTANT]
> **Quantization & NPU Compatibility Note:**
> This repository contains **Q4NX quantized weights** converted from [empero-ai/Qwable-9B-Claude-Fable-5](https://huggingface.co/empero-ai/Qwable-9B-Claude-Fable-5) to run natively on **FastFlowLM (`flm`) v1.0.3+** and **Lemonade** on AMD XDNA NPU hardware.
> 
> * **Model Type**: Quantized model conversion (NPU Q4NX format)
> * **Parent / Base Model**: [empero-ai/Qwable-9B-Claude-Fable-5](https://huggingface.co/empero-ai/Qwable-9B-Claude-Fable-5)
> * **Details**: Re-quantized to Q4NX format for FastFlowLM v1.0.3+ and Lemonade on AMD XDNA NPU. Fine-tuned for agentic coding and reasoning, configured with full EOS stop token sequence IDs ([248044, 248046]).
> * **Architecture**: Qwable 9B (Qwen3.5 9B architecture)
> * **Quantization Format**: Q4_K / Q4_1 / Q8_0 hybrid Q4NX
> * **Format**: `Q4NX` (safetensors format with AMD NPU block packing). Note that this is **not** a standard GGUF file; it is executed natively via `flm` / Lemonade on AMD Ryzen AI NPUs.

---

## Serving with Lemonade & FastFlowLM

To serve this model via Lemonade or FastFlowLM:

```bash
# Pull and run with FLM:
flm pull Qwable-9B-Claude-Fable-5-NPU2
flm serve Qwable-9B-Claude-Fable-5-NPU2 --ctx-len 32768 --port 8001
```

Or configure via Lemonade:
```bash
lemonade run Qwable-9B-Claude-Fable-5-NPU2
```

---

## Original Model Information (empero-ai/Qwable-9B-Claude-Fable-5)

Below is the model card from the upstream repository [empero-ai/Qwable-9B-Claude-Fable-5](https://huggingface.co/empero-ai/Qwable-9B-Claude-Fable-5):

---

<p align="center">
  <img src="qwable9b.jpg" alt="Qwable-9B-Claude-Fable-5" width="420"/>
</p>

# Qwable-9B-Claude-Fable-5

**Developed by [Empero](https://empero.org)**

Qwable-9B-Claude-Fable-5 is a full-parameter supervised fine-tune of
**[Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B)** on a curated mix of agentic coding and
reasoning traces. It is a distillation-style fine-tune: the training targets are outputs from other
assistants (Claude Fable 5 and a GPT-5.5 terminal agent), teaching the model to imitate their reasoning and
tool-use style on long, multi-turn coding and agent tasks.

> **Early release.** Qwable-9B-Claude-Fable-5 brings strong coding and agentic behavior out of the box. A
> full suite of quantitative benchmarks (coding, agentic, and safety) is underway and will be added to this
> card; training quality is already backed by held-out validation results (see [Evaluation](#evaluation)).
> See [Provenance & licensing](#provenance--licensing) for licensing notes.

## Model details

- **Developed by:** [Empero](https://empero.org)
- **Base model:** Qwen3.5-9B — a dense, natively **multimodal** model with a hybrid attention stack
  (3:1 Gated DeltaNet linear-attention to Gated full-attention), ~152k vocabulary, long native context.
- **Fine-tune type:** full parameter (all text-backbone weights trained). The **vision tower was frozen** —
  training was **text-only**, so vision behavior is inherited from the base and **was not tuned or tested**.
- **Objective:** supervised fine-tuning, **assistant-only loss** (the model is scored only on the
  assistant/completion tokens; prompts are masked out).
- **Languages:** primarily English.
- **License:** `apache-2.0`, inherited from the base weights — but see the data-provenance caveat below.

## Training data

| Source | Role | Approx. examples (after holdout) |
|---|---|---|
| [`Glint-Research/Fable-5-traces`](https://huggingface.co/datasets/Glint-Research/Fable-5-traces) | Claude Fable 5 reasoning + coding traces (`context` → `completion`) | ~4,585 |
| [`Roman1111111/gpt5.5-terminal`](https://huggingface.co/datasets/Roman1111111/gpt5.5-terminal) | GPT-5.5 terminal/agent task solutions (`system` + `prompt` → `solution`) | ~111 |

Both sources were normalized to a single chat format (`user`/`assistant`, with an optional `system` turn for
the terminal tasks) and concatenated. The natural mix is heavily skewed toward Fable traces (~97%); no
re-weighting was applied to the training set.

**Held-out eval split:** 100 examples were withheld from training — deliberately composed **80% Fable /
20% terminal** so the held-out loss carries signal on *both* task types rather than being dominated by Fable.

## Training procedure

Full-parameter supervised fine-tuning with [TRL](https://github.com/huggingface/trl), using:

- **Full-length traces, zero truncation** (`max_length = 76,800`) — even the longest multi-turn traces
  (~74k tokens) are trained in full.
- **Assistant-only loss** — the model is scored only on assistant/completion tokens; prompt tokens are masked.
- **Chunked cross-entropy** for memory-efficient long-context training.

| Hyperparameter | Value |
|---|---|
| Epochs | 2 |
| Effective batch size | 16 |
| Max sequence length | 76,800 (no truncation) |
| Learning rate | 1e-5 (cosine, 3% warmup) |
| Optimizer | AdamW (8-bit) |
| Precision | bf16 |
| Loss | chunked NLL, assistant-only |

## Evaluation

Training quality was tracked via **held-out validation loss and token-accuracy** on a 100-example split and
supplemented with a qualitative generation review (below). A full suite of **coding, agentic, and safety
benchmarks is in progress and will be published here.** Validation was run periodically during training:

| Step | eval loss | eval token-acc |
|---|---|---|
| 100 | 0.743 | 0.784 |
| 200 | 0.722 | 0.789 |
| 300 (≈ epoch 1) | 0.714 | 0.791 |
| 400 | 0.7135 | 0.791 |
| 500 | 0.713 | 0.791 |

**No overfitting observed.** Held-out loss decreased monotonically and then **plateaued (~0.71)** through the
second epoch — it never rose, even as train loss fell to ~0.64. Epoch-1 and final (epoch-2) checkpoints
generalize equivalently on held-out data.

> Note: `token-accuracy` is teacher-forced, per-token next-token accuracy over completion tokens only. It is
> **not** end-to-end correctness and tends to read high on consistent-style distillation data.

### Qualitative generation review

34 prompts spanning coding, terminal/agentic tasks, reasoning, explanation, instruction-following, and
honesty/calibration probes were run against the final checkpoint using Qwen3.5's recommended sampling
settings. Full unedited transcripts are in [`sample_generations.md`](sample_generations.md).

**Strengths.** Coding and terminal/agentic prompts were the strongest — correct, idiomatic solutions using
current tooling (e.g. `ss` over `netstat`, `git-filter-repo`, Argon2id) with security-aware judgment
(rotating a leaked key first, constant-time comparison, generic auth errors). Reasoning, instruction/format
following, and calibration probes were handled well. Roughly **27 of 34** responses were clean and correct.

The model is a **reasoning model**: every answer begins with a `<think>` block followed by the final
response — downstream consumers should parse out and strip the `<think>...</think>` span. See
[Limitations](#limitations) for usage tips.

## How to use

The base is a multimodal (image-text-to-text) architecture; for text-only use load it with
`AutoModelForImageTextToText`. Build the prompt with `tokenize=False` and then tokenize the string
(the recommended path for this tokenizer):

```python
import torch
from transformers import AutoModelForImageTextToText, AutoTokenizer

model_id = "empero-ai/Qwable-9B-Claude-Fable-5"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
    model_id, dtype="bfloat16", device_map="auto"
)

messages = [{"role": "user", "content": "Write a Python function that merges two sorted lists."}]
text = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tok(text, return_tensors="pt").to(model.device)

out = model.generate(
    **inputs, max_new_tokens=2048, do_sample=True,
    temperature=0.7, top_p=0.95, top_k=20, repetition_penalty=1.05,
)
# Output begins with a <think>...</think> reasoning block, then the final answer.
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```

`repetition_penalty=1.05` is a small deviation from Qwen's default (1.0) that prevents rare
non-terminating reasoning loops; allow generous `max_new_tokens` since the model reasons before answering.

**Requirements:** a recent `transformers` (Qwen3.5 support) plus the Gated DeltaNet kernels
(`flash-linear-attention` and a CUDA-matched `causal_conv1d` build) — without them the linear-attention
layers fall back to slow, memory-hungry PyTorch ops.

## Limitations

Qwable-9B-Claude-Fable-5 is a focused 9B model that shines on the coding, agentic, and reasoning tasks it was
trained for. A few characteristics are worth knowing to get the best out of it:

- **It's a reasoning model.** Each response opens with a `<think>` block before the final answer, so parse
  and strip the `<think>...</think>` span for end users. On open-ended or creative prompts it may reason at
  length — allow generous `max_new_tokens` and use `repetition_penalty≈1.05` (as in the snippet above) for
  consistently crisp completions.
- **Strongest within its domain.** Capability is concentrated in coding and agentic/tool-use tasks. For
  general-knowledge or long-form factual questions, treat specifics as you would any 9B model's — verify
  before relying on them, and don't expect knowledge of events outside the base model's training.
- **Reflects its base and teachers.** As a distillation fine-tune of Qwen3.5-9B on Claude Fable 5 and GPT-5.5
  traces, it carries the style and limits of those sources and received no extra safety tuning beyond the
  base model's. Add your own review/safety layer for production use.
- **Text-only fine-tune.** The base is multimodal, but only the text path was trained (vision left untouched
  and not evaluated here).

These are normal considerations for a compact, domain-focused model rather than blockers — used within its
wheelhouse with the sampling settings above, it's a capable and dependable coding/agentic assistant.

## Provenance & licensing

The model weights are released under **Apache-2.0**, inherited from the Qwen3.5-9B base. The fine-tuning data
comes from generated traces of Claude Fable 5 and GPT-5.5 (via the linked public datasets). Because those
traces originate from third-party assistants, the providers' terms may apply to downstream training and
distillation — so if you plan to build on this model commercially, it's worth confirming your use aligns with
those terms. Shared with the community for research and experimentation, as-is.

## Support / Donate

If this model helped you, consider supporting the project:

- **BTC**: `bc1qx6zepu6sfkvshgdmc4ewu6pk6rpadvpgffpp7v`
- **LTC**: `ltc1qv2mefzps2vtjcpwfx8xxdrpplrcvltswm68r7x`
- **XMR**: `42Dbm5xg5Nq26fdyzfEU7KBnAJfhi7Cvz5J2ex5CzHXkfKuNEJzYCcmJ1GTbgjFZ5MBx72sdG1G9239Cd6rsZfv4QeDkYJY`

## Acknowledgements

- Developed and released by [Empero](https://empero.org)
- Base model: [Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B) (Alibaba Qwen team)
- Datasets: [`Glint-Research/Fable-5-traces`](https://huggingface.co/datasets/Glint-Research/Fable-5-traces),
  [`Roman1111111/gpt5.5-terminal`](https://huggingface.co/datasets/Roman1111111/gpt5.5-terminal)
- Training: [TRL](https://github.com/huggingface/trl) + [Transformers](https://github.com/huggingface/transformers)