Upload _paper_results/reasoning_rl_multiseed_summary.md with huggingface_hub
Browse files
_paper_results/reasoning_rl_multiseed_summary.md
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Reasoning-RL multi-seed summary
|
| 2 |
+
|
| 3 |
+
Headline TF-grounding rate (TFG) across seeds for T1/T2/T3 reasoning grounded-RL (alpha=1.0, 250 steps, lr=5e-6, r=128). Higher TFG = more biologically grounded reasoning citations to expressed TFs.
|
| 4 |
+
|
| 5 |
+
| Task | n seeds | TFG mean ± std | TFG min, max | per-seed values |
|
| 6 |
+
|------|--------:|----------------|--------------|------------------|
|
| 7 |
+
| T1 | 3 | 0.4115 ± 0.0233 | [0.3964, 0.4384] | s=2: 0.3964, s=3: 0.3996, s=42: 0.4384 |
|
| 8 |
+
| T2 | 1 | 0.3650 ± 0.0000 | [0.3650, 0.3650] | s=42: 0.3650 |
|
| 9 |
+
| T3 | 3 | 0.2247 ± 0.0257 | [0.1951, 0.2397] | s=2: 0.2397, s=3: 0.1951, s=42: 0.2393 |
|
| 10 |
+
|
| 11 |
+
## T1 per-seed details
|
| 12 |
+
|
| 13 |
+
| seed | TFG | n_cited | n_grounded | n_halluc | reasoning_tags_rate |
|
| 14 |
+
|-----:|----:|--------:|-----------:|---------:|---------------------:|
|
| 15 |
+
| 2 | 0.3964 | 20.48 | 9.68 | 10.58 | 0.6000 |
|
| 16 |
+
| 3 | 0.3996 | 22.84 | 13.74 | 8.76 | 0.2000 |
|
| 17 |
+
| 42 | 0.4384 | 27.46 | 14.58 | 12.36 | 0.1200 |
|
| 18 |
+
|
| 19 |
+
## T2 per-seed details
|
| 20 |
+
|
| 21 |
+
| seed | TFG | n_cited | n_grounded | n_halluc | reasoning_tags_rate |
|
| 22 |
+
|-----:|----:|--------:|-----------:|---------:|---------------------:|
|
| 23 |
+
| 42 | 0.3650 | 20.58 | 9.84 | 10.08 | 0.4400 |
|
| 24 |
+
|
| 25 |
+
## T3 per-seed details
|
| 26 |
+
|
| 27 |
+
| seed | TFG | n_cited | n_grounded | n_halluc | reasoning_tags_rate |
|
| 28 |
+
|-----:|----:|--------:|-----------:|---------:|---------------------:|
|
| 29 |
+
| 2 | 0.2397 | 14.38 | 6.46 | 7.52 | 0.6400 |
|
| 30 |
+
| 3 | 0.1951 | 15.08 | 5.22 | 9.54 | 0.8000 |
|
| 31 |
+
| 42 | 0.2393 | 20.06 | 7.80 | 11.88 | 0.6400 |
|