---
base_model: Qwen/Qwen3-1.7B
library_name: transformers
pipeline_tag: text-generation
language:
- en
tags:
- kanha
- qwen3
- qlora
- mlc
---

# Kanha Qwen3 experiment

## Run identity

- Run ID: `incidentio-incidents-1.7b-qlora-v1`
- Base model: `Qwen/Qwen3-1.7B`
- Base model revision: `70d244cc86ccca08cf5af4e1e306ecf908b1ad5e`
- Tokenizer revision: `70d244cc86ccca08cf5af4e1e306ecf908b1ad5e`
- Training method: `qlora`
- Final merged dtype: `bfloat16`
- Source site: https://docs.incident.io
- Dataset hash: `60d31b62ef843cac4863ffbed1c16127dcc53899c11f137bd802df4191804a94`
- Train split: 276 records (`c9f3c08e10208c56db651f2df9341c18c6e87f8c89fbfe1675f3eee9ec044659`)
- Validation split: 33 records (`eb14d70e7e0b965785778acea21669327bb0d7a13c8367356876afc0e8a714c4`)
- Holdout split: 0 records (`e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855`)

## Hyperparameters

- Maximum sequence length: 2048
- Seed: 42
- Epochs: 4.0
- Learning rate: 5e-05
- Per-device batch size: 4
- Gradient accumulation steps: 2
- Warmup ratio: 0.1
- Assistant-only loss: true
- LoRA rank: 64
- LoRA alpha: 32
- LoRA dropout: 0.05
- LoRA targets: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj

## Evaluation

- dates_recall: `1.0`
- deterministic_pass_rate: `0.0`
- list_recall: `1.0`
- numbers_recall: `1.0`
- refusal_rate: `0.0`
- requires_review_rate: `1.0`
- total: `3`
- unsupported_value_rate: `0.0`
- urls_recall: `1.0`

Deterministic scoring and the server-side Transformers benchmark are not browser qualification. Validate the exact converted model in the target browser and device environments.

## MLC availability

MLC artifacts using `q4f16_1` quantization are available under `mlc/`.

## Intended use

This checkpoint is intended for research comparing training methods on the same Kanha website-derived dataset and for controlled evaluation of website question answering.

## Limitations

The checkpoint can produce incorrect, incomplete, or stale answers. It may memorize training content. Review outputs, test representative failure cases, and qualify the exact runtime before any user-facing use.

## Provenance artifacts

- `research/run-manifest.json`
- `research/training-config.yaml`
- `research/publication-inventory.json`
- `research/conversion-manifest.json`
- `research/evaluation/metrics.json`
- `research/evaluation/evaluation-manifest.json`