Benak release
Browse files- README.md +1 -1
- code/build.py +1 -1
- data/sea_thinking.jsonl +0 -0
- results.json +1 -1
README.md
CHANGED
|
@@ -57,7 +57,7 @@ configs:
|
|
| 57 |
|---|---|---|
|
| 58 |
| `atlas_runs` | 3,200 | Belebele, 19 languages × 40 parallel items × 4 prompting conditions (+ effort=high subset). Every call is included, with full reasoning, answer, grade, token usage and GlotLID profile. |
|
| 59 |
| `math_runs` | 1,276 | 58 code-verified math word problems × 9 languages × {C0, C2}, plus C3 (the `benak_lang` wrapper) for Indonesian, Javanese, Sundanese and Thai. |
|
| 60 |
-
| `sea_thinking` | 1,
|
| 61 |
| `replication_profiled` | 4,139 | Beam traces that other entrants had already published (neulketing Korean, esscor TamilBench), re-scored with our detector. These cost $0 of inference. |
|
| 62 |
| `grammar_book_bugis.json` | 20 calls | Can a book in Beam's 1M-token context teach it Buginese? Includes hypotheses, chrF++ scores and the reasoning-language probe. |
|
| 63 |
|
|
|
|
| 57 |
|---|---|---|
|
| 58 |
| `atlas_runs` | 3,200 | Belebele, 19 languages × 40 parallel items × 4 prompting conditions (+ effort=high subset). Every call is included, with full reasoning, answer, grade, token usage and GlotLID profile. |
|
| 59 |
| `math_runs` | 1,276 | 58 code-verified math word problems × 9 languages × {C0, C2}, plus C3 (the `benak_lang` wrapper) for Indonesian, Javanese, Sundanese and Thai. |
|
| 60 |
+
| `sea_thinking` | 1,204 | **The fix.** Correct answers whose reasoning is ≥ 80% in the target language (18 non-English languages; most rows are Southeast Asian), in the [HuggingFaceH4/Multilingual-Thinking](https://huggingface.co/datasets/HuggingFaceH4/Multilingual-Thinking) schema (`reasoning_language, developer, user, analysis, final, messages`). That is the format Reflection's own TRL recipe uses, so it can be used directly for SFT. `source` marks native traces vs Beam-rewritten traces. |
|
| 61 |
| `replication_profiled` | 4,139 | Beam traces that other entrants had already published (neulketing Korean, esscor TamilBench), re-scored with our detector. These cost $0 of inference. |
|
| 62 |
| `grammar_book_bugis.json` | 20 calls | Can a book in Beam's 1M-token context teach it Buginese? Includes hypotheses, chrF++ scores and the reasoning-language probe. |
|
| 63 |
|
code/build.py
CHANGED
|
@@ -35,7 +35,7 @@ think = []
|
|
| 35 |
_bel = {}
|
| 36 |
for r in math_ + atlas + rew:
|
| 37 |
ts = (r.get("lid") or {}).get("target_share")
|
| 38 |
-
if r.get("correct") and ts is not None and ts >= 0.8 and r.get("content"):
|
| 39 |
lang = r["lang"]
|
| 40 |
if "link" in r: # Belebele item: rebuild the exact prompt Beam saw (C0-C3 share the same user turn)
|
| 41 |
from atlas import mc_prompt
|
|
|
|
| 35 |
_bel = {}
|
| 36 |
for r in math_ + atlas + rew:
|
| 37 |
ts = (r.get("lid") or {}).get("target_share")
|
| 38 |
+
if r.get("correct") and ts is not None and ts >= 0.8 and r.get("content") and r["lang"] != "eng_Latn":
|
| 39 |
lang = r["lang"]
|
| 40 |
if "link" in r: # Belebele item: rebuild the exact prompt Beam saw (C0-C3 share the same user turn)
|
| 41 |
from atlas import mc_prompt
|
data/sea_thinking.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results.json
CHANGED
|
@@ -2814,5 +2814,5 @@
|
|
| 2814 |
"huggingface": 2116
|
| 2815 |
},
|
| 2816 |
"cost_usd": 2.8987,
|
| 2817 |
-
"sea_thinking_rows":
|
| 2818 |
}
|
|
|
|
| 2814 |
"huggingface": 2116
|
| 2815 |
},
|
| 2816 |
"cost_usd": 2.8987,
|
| 2817 |
+
"sea_thinking_rows": 1204
|
| 2818 |
}
|