Jnx03's picture
v14a-step160: sft-lora
529fc0b verified
|
Raw
History Blame Contribute Delete
1.28 kB
---
library_name: peft
base_model: typhoon-ai/typhoon-s-thaillm-8b-instruct-research-preview
license: apache-2.0
tags:
- thai
- kanitakorn
- single-model
- greedy-eval
language:
- th
- en
---
# Jnx03/kanitakorn-v14a-v14a-step160
Kanitakorn 2026-06-13 campaign artifact.
- Base: `typhoon-ai/typhoon-s-thaillm-8b-instruct-research-preview`
- Version: `v14a-step160`
- Method: `sft-lora`
- Final-claim policy: one model, no BoN, no self-consistency, no ensemble, no model routing.
- Date: 2026-06-13
## Scores
| Benchmark | Score | Target | Status |
|---|---:|---:|---|
| aime24_th | pending | >15.0 | pending |
| aime24 | pending | >25.0 | pending |
| math500_th | pending | >56.0 | pending |
| math500 | pending | >82.0 | pending |
| livecodebench_th | pending | >35.0 | pending |
| livecodebench | pending | >60.0 | pending |
| openthaieval | pending | >80.0 | pending |
| hotpotqa_th_en | pending | >46.0 | pending |
| instruction_following_th_en | pending | >57.0 | pending |
| mt_bench_th_en | pending | >85.0 | pending |
| thaiexam | pending | >70.0 | pending |
| ifeval_th | pending | >82.0 | pending |
## Notes
Automatic checkpoint publish from runs/20260613_typhoon8b_v14a_harderror_fast_r16_lr5e6_400/checkpoint-160. Single-model artifact; no BoN/self-consistency.