Complete the evaluation table (15 checkpoints, 20000-22000 added)

README.md

@@ -20,9 +20,8 @@ Decoder-only language model, 1 073 912 851 parameters, Polish-dominant. **Slayer
 Slayer1B 1.06** from its final checkpoint (step 13848) on 8 891 005 707 training tokens (about 7.0 B new and 1.9 B
 replayed from the 1.06 data), at the same constant learning rate and global batch.
 
-**Status: training complete at step 22327 (6 November 2026).** The training-time evaluation table below includes the
-final checkpoint (22327); the checkpoints between 19000 and 22327 (20000, 21000, 21450, 22000) are being evaluated
-with the same protocol, and each result is added here when it is ready. A 1.07 model release (converted weights) has not been
+**Status: training complete at step 22327 (6 November 2026).** All 15 evaluated checkpoints of the run, from the 1.06
+starting point (13848) to the final checkpoint (22327), are in the training-time evaluation table below. A 1.07 model release (converted weights) has not been
 published in this repository yet. Native training checkpoints (raw PyTorch training state `ckpt_N.pt`, not a
 ready-to-use model) are published file by file in
 [SlayerLab/Slayer1B-1.07-checkpoints](https://huggingface.co/SlayerLab/Slayer1B-1.07-checkpoints), with sha256 sums.
@@ -77,24 +76,34 @@ whose exact normalized text was already in the 1.06 English data were excluded.
 ## Training-time evaluation
 
 Same automatic evaluator and the same zero-shot protocol as the 1.06 card (BLiMP, ARC-Easy, WikiText-2, MultiBLiMP-PL,
-ARC-Easy-PL eu20). 1.07 checkpoints are evaluated on CPU; the 1.06 checkpoint (step 13848) evaluated on the same CPU
-setup matches the 1.06 card values at card rounding. These are **not** the lm-evaluation-harness benchmark used in the
-Results table of 1.00–1.03.
-
-| metric | 1.06 (13848) | 13998 | 14000 | 15000 | 15730 | 16000 | 17000 | 18000 | 18590 | 19000 | **22327 (final)** |
-|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
-| BLiMP (EN), accuracy % | 76.12 | 76.29 | 76.33 | 77.94 | 77.32 | 77.13 | 77.61 | 77.88 | 77.46 | 77.74 | **77.90** |
-| ARC-Easy (EN), accuracy % | 52.99 | 54.34 | 54.21 | 54.59 | 55.64 | 56.73 | 56.48 | 56.52 | 56.52 | 56.65 | **58.67** |
-| WikiText-2, byte perplexity ↓ | 2.2849 | 2.2618 | 2.2601 | 2.2321 | 2.2237 | 2.2205 | 2.2138 | 2.2042 | 2.2145 | 2.2029 | **2.1837** |
-| WikiText-2, token perplexity ↓ | 30.03 | 28.80 | 28.71 | 27.28 | 26.85 | 26.70 | 26.37 | 25.90 | 26.40 | 25.84 | **24.92** |
-| MultiBLiMP (PL), accuracy % | 98.41 | 98.66 | 98.59 | 98.62 | 98.62 | 98.81 | 98.87 | 98.84 | 98.62 | 98.53 | **98.84** |
-| ARC-Easy (PL, eu20), accuracy % | 45.03 | 45.37 | 44.91 | 45.92 | 45.75 | 46.25 | 45.41 | 46.21 | 47.14 | 46.80 | **46.93** |
+ARC-Easy-PL eu20). Checkpoints 13848–19000 and 22327 were evaluated on CPU, checkpoints 20000–22000 on an NVIDIA RTX 4090
+(rented). On both devices the 1.06 checkpoint (step 13848) reproduces the 1.06 card values at card rounding, and on the
+GPU the published Slayer1B 1.02 also reproduces its card values, so all rows are on the same scale. These are **not** the
+lm-evaluation-harness benchmark used in the Results table of 1.00–1.03.
+
+| step | BLiMP (EN), % | ARC-Easy (EN), % | WikiText-2 byte PPL ↓ | WikiText-2 token PPL ↓ | MultiBLiMP (PL), % | ARC-Easy (PL, eu20), % | device |
+|---:|---:|---:|---:|---:|---:|---:|---|
+| 13848 (1.06) | 76.12 | 52.99 | 2.2849 | 30.03 | 98.41 | 45.03 | CPU |
+| 13998 | 76.29 | 54.34 | 2.2618 | 28.80 | 98.66 | 45.37 | CPU |
+| 14000 | 76.33 | 54.21 | 2.2601 | 28.71 | 98.59 | 44.91 | CPU |
+| 15000 | 77.94 | 54.59 | 2.2321 | 27.28 | 98.62 | 45.92 | CPU |
+| 15730 | 77.32 | 55.64 | 2.2237 | 26.85 | 98.62 | 45.75 | CPU |
+| 16000 | 77.13 | 56.73 | 2.2205 | 26.70 | 98.81 | 46.25 | CPU |
+| 17000 | 77.61 | 56.48 | 2.2138 | 26.37 | 98.87 | 45.41 | CPU |
+| 18000 | 77.88 | 56.52 | 2.2042 | 25.90 | 98.84 | 46.21 | CPU |
+| 18590 | 77.46 | 56.52 | 2.2145 | 26.40 | 98.62 | 47.14 | CPU |
+| 19000 | 77.74 | 56.65 | 2.2029 | 25.84 | 98.53 | 46.80 | CPU |
+| 20000 | 77.68 | 56.40 | 2.1940 | 25.41 | 98.66 | 46.55 | RTX 4090 |
+| 21000 | 77.53 | 56.69 | 2.1813 | 24.81 | 98.62 | 47.01 | RTX 4090 |
+| 21450 | 77.44 | 58.29 | 2.1867 | 25.06 | 98.75 | 47.31 | RTX 4090 |
+| 22000 | 77.16 | 57.28 | 2.1794 | 24.72 | 98.75 | 46.97 | RTX 4090 |
+| **22327 (final)** | **77.90** | **58.67** | **2.1837** | **24.92** | **98.84** | **46.93** | CPU |
 
 ARC-Easy has 2 376 questions per language, so differences below about one percentage point between neighbouring
-checkpoints are within noise. WikiText-2 byte perplexity at 18590 was 0.010 higher than at 18000. The final checkpoint
-(22327) has the lowest WikiText-2 perplexity and the highest English ARC-Easy accuracy of the run (+2.0 points over
-19000, together with small gains on all other metrics); the checkpoints in between are not measured yet, so the chart
-joins 19000 and 22327 directly. Each point is a single measurement of a constant-learning-rate run. The training loss in the figure is measured
+checkpoints are within noise. From 20000 on, WikiText-2 byte perplexity moves within about ±0.005 around 2.18 instead of
+falling steadily (lowest 2.1794 at 22000; 2.1837 at the final step): the run used a constant learning rate without decay,
+so the last checkpoint is where training stopped, not a converged minimum. The final checkpoint has the highest English
+ARC-Easy accuracy of the run (58.67, +2.0 points over 19000). Each point is a single measurement. The training loss in the figure is measured
 on the 1.07 data mixture (44 % English), so it is not directly comparable with the training loss of 1.00–1.06.
 
 The benchmark caveat of the 1.03 card applies to the 1.00–1.06 part of the training. For 1.07, all nine data files