dawidmajewski commited on
Commit
b9abe2f
·
verified ·
1 Parent(s): a2e53d9

Finalize Pollock 1.4 release metadata

Browse files
README.md CHANGED
@@ -127,7 +127,7 @@ Wspólny zestaw [`fixed-sampling-v1`](./inference-samples/README.md) pokazuje te
127
 
128
  | Rewizja | Wydanie | Wagi | Historia |
129
  |---|---|---|---|
130
- | **r006** | **Pollock 1.4** | PR commit pending | [`r006.md`](./training-history/r006.md) |
131
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | [`r005.md`](./training-history/r005.md) |
132
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | [`r004.md`](./training-history/r004.md) |
133
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | [`r003.md`](./training-history/r003.md) |
@@ -216,7 +216,7 @@ The shared [`fixed-sampling-v1`](./inference-samples/README.md) suite runs the s
216
 
217
  | Revision | Release | Weights | History |
218
  |---|---|---|---|
219
- | **r006** | **Pollock 1.4** | PR commit pending | [`r006.md`](./training-history/r006.md) |
220
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | [`r005.md`](./training-history/r005.md) |
221
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | [`r004.md`](./training-history/r004.md) |
222
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | [`r003.md`](./training-history/r003.md) |
 
127
 
128
  | Rewizja | Wydanie | Wagi | Historia |
129
  |---|---|---|---|
130
+ | **r006** | **Pollock 1.4** | [`a2e53d9`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9) | [`r006.md`](./training-history/r006.md) |
131
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | [`r005.md`](./training-history/r005.md) |
132
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | [`r004.md`](./training-history/r004.md) |
133
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | [`r003.md`](./training-history/r003.md) |
 
216
 
217
  | Revision | Release | Weights | History |
218
  |---|---|---|---|
219
+ | **r006** | **Pollock 1.4** | [`a2e53d9`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9) | [`r006.md`](./training-history/r006.md) |
220
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | [`r005.md`](./training-history/r005.md) |
221
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | [`r004.md`](./training-history/r004.md) |
222
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | [`r003.md`](./training-history/r003.md) |
inference-samples/README.md CHANGED
@@ -23,6 +23,7 @@ Exact token replay was verified in the recorded environment. Sampling may diverg
23
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | `3a9964a0ba5f6d20ef6061d2a7e4442c1b001209ebe80f7b7148a25573decaa9` |
24
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | `3c453f22d4bb70228e5cca79e183a425e0f0221e54bd782d12027c42c943c880` |
25
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | `3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c` |
 
26
 
27
  ## Results
28
 
@@ -110,6 +111,15 @@ social
110
 
111
  Output SHA-256: `ed9cdec69775563f011b56008d68f9f49f6aa8b74a23e806a8a1e2f7f333557c`
112
 
 
 
 
 
 
 
 
 
 
113
  ### Explanation
114
 
115
  Prompt: `Photosynthesis is the process by which`
@@ -175,6 +185,18 @@ Carbon dioxide is a gas, meaning it contains a trace amount of energy. Carbon di
175
 
176
  Output SHA-256: `1f68683a0156d242a283349bdc053bb92d0d41a1a6c94daf5cfad67b88fb23d5`
177
 
 
 
 
 
 
 
 
 
 
 
 
 
178
  ### Story
179
 
180
  Prompt: `In a distant future, humanity discovered`
@@ -244,6 +266,16 @@ The ability to read and write at the same time was something the world
244
 
245
  Output SHA-256: `7feee4648e0b22f7fdf4365efea4ac48c21fd9e1d87afc7c46825ff11808e198`
246
 
 
 
 
 
 
 
 
 
 
 
247
  ### Code
248
 
249
  Prompt: `def fibonacci(n):`
@@ -354,6 +386,28 @@ while True
354
 
355
  Output SHA-256: `5767898a26ba2104eb5ede9f29195d39715573f09d9d415f58881e5000cd688f`
356
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
357
  ## Reproduce or verify
358
 
359
  Run from this directory. The check downloads approximately 2.4 GB of published model artifacts if they are not already cached.
 
23
  | r003 | Pollock 1.1 | [`698984b`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/698984b1d1b96c9b6ffaff00c7fc2e78e140e842) | `3a9964a0ba5f6d20ef6061d2a7e4442c1b001209ebe80f7b7148a25573decaa9` |
24
  | r004 | Pollock 1.2 | [`30feb81`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/30feb81e5097eef13b939e1827d98e0458bf602d) | `3c453f22d4bb70228e5cca79e183a425e0f0221e54bd782d12027c42c943c880` |
25
  | r005 | Pollock 1.3 | [`e780025`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/e780025f15e06bc3765a74d973906eb8a11c022c) | `3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c` |
26
+ | r006 | Pollock 1.4 | [`a2e53d9`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9) | `8e74aa4d34464229e86a1831fb34d442f91e22aa600e0cf2503891957fdc6a43` |
27
 
28
  ## Results
29
 
 
111
 
112
  Output SHA-256: `ed9cdec69775563f011b56008d68f9f49f6aa8b74a23e806a8a1e2f7f333557c`
113
 
114
+ #### r006 - Pollock 1.4
115
+
116
+ ````text
117
+ The capital of France is the capital city of England. It is on the banks of the River Liguria, and is divided into four provinces. Its capital is Paris, and is situated at the end of the Aisne River.
118
+ The municipality of Aisne is situated on the banks of the Neuve-Brieu river, which flows through the north of France. The capital of the city is Aisne-sur-Ligne, on the banks of
119
+ ````
120
+
121
+ Output SHA-256: `b47fcb42354b8f735c7b98cfdf93605c6e095ed8290e2ddb3329886df244e99f`
122
+
123
  ### Explanation
124
 
125
  Prompt: `Photosynthesis is the process by which`
 
185
 
186
  Output SHA-256: `1f68683a0156d242a283349bdc053bb92d0d41a1a6c94daf5cfad67b88fb23d5`
187
 
188
+ #### r006 - Pollock 1.4
189
+
190
+ ````text
191
+ Photosynthesis is the process by which plants and animals obtain oxygen from the air they breathe.
192
+ Chemical change: A chemical reaction is the change in state of matter or substance that occurs in a substance.
193
+ Chemical reaction: The combining of two substances to produce a new substance or a new compound.
194
+ Chemical reaction: The chemical reaction between two substances to give a new compound.
195
+ Carbon gas: A mixture of gases in which carbon dioxide is replaced by a gas; in which carbon dioxide
196
+ ````
197
+
198
+ Output SHA-256: `c09d560713828851410d37fcd7c648d49c2bedf6e9c8325bb47cb32a8015bfd4`
199
+
200
  ### Story
201
 
202
  Prompt: `In a distant future, humanity discovered`
 
266
 
267
  Output SHA-256: `7feee4648e0b22f7fdf4365efea4ac48c21fd9e1d87afc7c46825ff11808e198`
268
 
269
+ #### r006 - Pollock 1.4
270
+
271
+ ````text
272
+ In a distant future, humanity discovered that the human race is not only a highly advanced civilization, but also a civilization founded on a shared philosophy that believes in the power of living and the dignity of the earth. The concept of humanity transcends the limits of human understanding, but the human spirit is built on a complex interplay of human intelligence and cultural heritage.
273
+ user
274
+ This is fascinating! Can you tell me more about the specific cultural values that have shaped the concept of humanity?
275
+ ````
276
+
277
+ Output SHA-256: `6455cdbe15a71ae11ff476739071f31b3f86b8bf4d97f3dabe99a7d1595c5c7c`
278
+
279
  ### Code
280
 
281
  Prompt: `def fibonacci(n):`
 
386
 
387
  Output SHA-256: `5767898a26ba2104eb5ede9f29195d39715573f09d9d415f58881e5000cd688f`
388
 
389
+ #### r006 - Pollock 1.4
390
+
391
+ ````text
392
+ def fibonacci(n):
393
+ return n * fibonacci(n)
394
+ # if n > 1:
395
+ # return n + fibonacci(n - 1)
396
+ # return fibonacci(n)
397
+ #
398
+ # return fibonacci(2)
399
+ #
400
+ # return fibonacci(4)
401
+ #
402
+ # return fibonacci(6)
403
+ #
404
+ # return fibonacci(10)
405
+ #
406
+
407
+ ````
408
+
409
+ Output SHA-256: `ad3aa2ea2cab9a46ae5f52a9cc109bcf5f7362e1d214ceb791ee2dd6995a345f`
410
+
411
  ## Reproduce or verify
412
 
413
  Run from this directory. The check downloads approximately 2.4 GB of published model artifacts if they are not already cached.
inference-samples/config.json CHANGED
@@ -54,6 +54,12 @@
54
  "release": "Pollock 1.3",
55
  "commit": "e780025f15e06bc3765a74d973906eb8a11c022c",
56
  "model_sha256": "3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c"
 
 
 
 
 
 
57
  }
58
  ],
59
  "prompts": [
 
54
  "release": "Pollock 1.3",
55
  "commit": "e780025f15e06bc3765a74d973906eb8a11c022c",
56
  "model_sha256": "3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c"
57
+ },
58
+ {
59
+ "revision_id": "r006",
60
+ "release": "Pollock 1.4",
61
+ "commit": "a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9",
62
+ "model_sha256": "8e74aa4d34464229e86a1831fb34d442f91e22aa600e0cf2503891957fdc6a43"
63
  }
64
  ],
65
  "prompts": [
inference-samples/results.json CHANGED
@@ -2454,6 +2454,497 @@
2454
  "full_text_sha256": "5767898a26ba2104eb5ede9f29195d39715573f09d9d415f58881e5000cd688f"
2455
  }
2456
  ]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2457
  }
2458
  ]
2459
  }
 
2454
  "full_text_sha256": "5767898a26ba2104eb5ede9f29195d39715573f09d9d415f58881e5000cd688f"
2455
  }
2456
  ]
2457
+ },
2458
+ {
2459
+ "revision_id": "r006",
2460
+ "release": "Pollock 1.4",
2461
+ "commit": "a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9",
2462
+ "model_sha256": "8e74aa4d34464229e86a1831fb34d442f91e22aa600e0cf2503891957fdc6a43",
2463
+ "loaded_model_version": "1.4",
2464
+ "results": [
2465
+ {
2466
+ "prompt_id": "general_knowledge",
2467
+ "prompt": "The capital of France is",
2468
+ "prompt_token_ids": [
2469
+ 492,
2470
+ 4514,
2471
+ 286,
2472
+ 4413,
2473
+ 311
2474
+ ],
2475
+ "completion_token_ids": [
2476
+ 263,
2477
+ 4514,
2478
+ 2329,
2479
+ 286,
2480
+ 3717,
2481
+ 46,
2482
+ 665,
2483
+ 311,
2484
+ 336,
2485
+ 263,
2486
+ 8700,
2487
+ 286,
2488
+ 263,
2489
+ 4286,
2490
+ 440,
2491
+ 325,
2492
+ 329,
2493
+ 572,
2494
+ 44,
2495
+ 292,
2496
+ 311,
2497
+ 5664,
2498
+ 685,
2499
+ 1689,
2500
+ 881,
2501
+ 260,
2502
+ 633,
2503
+ 46,
2504
+ 5646,
2505
+ 4514,
2506
+ 311,
2507
+ 6703,
2508
+ 44,
2509
+ 292,
2510
+ 311,
2511
+ 1640,
2512
+ 5446,
2513
+ 405,
2514
+ 263,
2515
+ 1009,
2516
+ 286,
2517
+ 263,
2518
+ 335,
2519
+ 271,
2520
+ 618,
2521
+ 4286,
2522
+ 46,
2523
+ 10,
2524
+ 492,
2525
+ 10715,
2526
+ 1523,
2527
+ 286,
2528
+ 335,
2529
+ 271,
2530
+ 618,
2531
+ 311,
2532
+ 1640,
2533
+ 5446,
2534
+ 336,
2535
+ 263,
2536
+ 8700,
2537
+ 286,
2538
+ 263,
2539
+ 2775,
2540
+ 117,
2541
+ 314,
2542
+ 45,
2543
+ 66,
2544
+ 3388,
2545
+ 117,
2546
+ 4952,
2547
+ 44,
2548
+ 506,
2549
+ 9271,
2550
+ 857,
2551
+ 263,
2552
+ 4097,
2553
+ 286,
2554
+ 4413,
2555
+ 46,
2556
+ 382,
2557
+ 4514,
2558
+ 286,
2559
+ 263,
2560
+ 2329,
2561
+ 311,
2562
+ 335,
2563
+ 271,
2564
+ 618,
2565
+ 45,
2566
+ 9745,
2567
+ 45,
2568
+ 76,
2569
+ 579,
2570
+ 101,
2571
+ 44,
2572
+ 336,
2573
+ 263,
2574
+ 8700,
2575
+ 286
2576
+ ],
2577
+ "completion_tokens": 100,
2578
+ "stopped_on_eos": false,
2579
+ "completion": " the capital city of England. It is on the banks of the River Liguria, and is divided into four provinces. Its capital is Paris, and is situated at the end of the Aisne River.\nThe municipality of Aisne is situated on the banks of the Neuve-Brieu river, which flows through the north of France. The capital of the city is Aisne-sur-Ligne, on the banks of",
2580
+ "full_text": "The capital of France is the capital city of England. It is on the banks of the River Liguria, and is divided into four provinces. Its capital is Paris, and is situated at the end of the Aisne River.\nThe municipality of Aisne is situated on the banks of the Neuve-Brieu river, which flows through the north of France. The capital of the city is Aisne-sur-Ligne, on the banks of",
2581
+ "full_text_sha256": "b47fcb42354b8f735c7b98cfdf93605c6e095ed8290e2ddb3329886df244e99f"
2582
+ },
2583
+ {
2584
+ "prompt_id": "explanation",
2585
+ "prompt": "Photosynthesis is the process by which",
2586
+ "prompt_token_ids": [
2587
+ 3712,
2588
+ 316,
2589
+ 394,
2590
+ 11580,
2591
+ 8310,
2592
+ 311,
2593
+ 263,
2594
+ 1170,
2595
+ 411,
2596
+ 506
2597
+ ],
2598
+ "completion_token_ids": [
2599
+ 3335,
2600
+ 292,
2601
+ 3620,
2602
+ 2663,
2603
+ 5715,
2604
+ 433,
2605
+ 263,
2606
+ 1773,
2607
+ 548,
2608
+ 280,
2609
+ 674,
2610
+ 258,
2611
+ 46,
2612
+ 10,
2613
+ 67,
2614
+ 1601,
2615
+ 481,
2616
+ 1477,
2617
+ 58,
2618
+ 335,
2619
+ 4144,
2620
+ 5372,
2621
+ 311,
2622
+ 263,
2623
+ 1477,
2624
+ 287,
2625
+ 1158,
2626
+ 286,
2627
+ 2578,
2628
+ 378,
2629
+ 8562,
2630
+ 342,
2631
+ 4682,
2632
+ 287,
2633
+ 257,
2634
+ 8562,
2635
+ 46,
2636
+ 10,
2637
+ 67,
2638
+ 1601,
2639
+ 481,
2640
+ 5372,
2641
+ 58,
2642
+ 382,
2643
+ 1984,
2644
+ 1806,
2645
+ 286,
2646
+ 734,
2647
+ 11163,
2648
+ 290,
2649
+ 3371,
2650
+ 257,
2651
+ 712,
2652
+ 8562,
2653
+ 378,
2654
+ 257,
2655
+ 712,
2656
+ 9715,
2657
+ 46,
2658
+ 10,
2659
+ 67,
2660
+ 1601,
2661
+ 481,
2662
+ 5372,
2663
+ 58,
2664
+ 382,
2665
+ 4144,
2666
+ 5372,
2667
+ 907,
2668
+ 734,
2669
+ 11163,
2670
+ 290,
2671
+ 1775,
2672
+ 257,
2673
+ 712,
2674
+ 9715,
2675
+ 46,
2676
+ 10,
2677
+ 7658,
2678
+ 2743,
2679
+ 2924,
2680
+ 58,
2681
+ 335,
2682
+ 7910,
2683
+ 286,
2684
+ 10458,
2685
+ 287,
2686
+ 506,
2687
+ 3781,
2688
+ 8726,
2689
+ 311,
2690
+ 6713,
2691
+ 411,
2692
+ 257,
2693
+ 2924,
2694
+ 59,
2695
+ 287,
2696
+ 506,
2697
+ 3781,
2698
+ 8726
2699
+ ],
2700
+ "completion_tokens": 100,
2701
+ "stopped_on_eos": false,
2702
+ "completion": " plants and animals obtain oxygen from the air they breathe.\nChemical change: A chemical reaction is the change in state of matter or substance that occurs in a substance.\nChemical reaction: The combining of two substances to produce a new substance or a new compound.\nChemical reaction: The chemical reaction between two substances to give a new compound.\nCarbon gas: A mixture of gases in which carbon dioxide is replaced by a gas; in which carbon dioxide",
2703
+ "full_text": "Photosynthesis is the process by which plants and animals obtain oxygen from the air they breathe.\nChemical change: A chemical reaction is the change in state of matter or substance that occurs in a substance.\nChemical reaction: The combining of two substances to produce a new substance or a new compound.\nChemical reaction: The chemical reaction between two substances to give a new compound.\nCarbon gas: A mixture of gases in which carbon dioxide is replaced by a gas; in which carbon dioxide",
2704
+ "full_text_sha256": "c09d560713828851410d37fcd7c648d49c2bedf6e9c8325bb47cb32a8015bfd4"
2705
+ },
2706
+ {
2707
+ "prompt_id": "story",
2708
+ "prompt": "In a distant future, humanity discovered",
2709
+ "prompt_token_ids": [
2710
+ 750,
2711
+ 257,
2712
+ 10103,
2713
+ 2629,
2714
+ 44,
2715
+ 12187,
2716
+ 4850
2717
+ ],
2718
+ "completion_token_ids": [
2719
+ 342,
2720
+ 263,
2721
+ 1811,
2722
+ 4909,
2723
+ 311,
2724
+ 434,
2725
+ 766,
2726
+ 257,
2727
+ 4295,
2728
+ 5787,
2729
+ 4161,
2730
+ 1353,
2731
+ 44,
2732
+ 520,
2733
+ 638,
2734
+ 257,
2735
+ 4161,
2736
+ 1353,
2737
+ 7887,
2738
+ 336,
2739
+ 257,
2740
+ 5354,
2741
+ 10102,
2742
+ 342,
2743
+ 11980,
2744
+ 287,
2745
+ 263,
2746
+ 1319,
2747
+ 286,
2748
+ 3099,
2749
+ 292,
2750
+ 263,
2751
+ 289,
2752
+ 579,
2753
+ 420,
2754
+ 286,
2755
+ 263,
2756
+ 3411,
2757
+ 46,
2758
+ 382,
2759
+ 2713,
2760
+ 286,
2761
+ 12187,
2762
+ 11261,
2763
+ 1628,
2764
+ 263,
2765
+ 7031,
2766
+ 286,
2767
+ 1811,
2768
+ 3137,
2769
+ 44,
2770
+ 520,
2771
+ 263,
2772
+ 1811,
2773
+ 4042,
2774
+ 311,
2775
+ 3063,
2776
+ 336,
2777
+ 257,
2778
+ 2481,
2779
+ 910,
2780
+ 1617,
2781
+ 286,
2782
+ 1811,
2783
+ 8096,
2784
+ 292,
2785
+ 5289,
2786
+ 699,
2787
+ 9595,
2788
+ 46,
2789
+ 12287,
2790
+ 10,
2791
+ 12286,
2792
+ 1512,
2793
+ 10,
2794
+ 1336,
2795
+ 311,
2796
+ 9311,
2797
+ 7152,
2798
+ 33,
2799
+ 1885,
2800
+ 360,
2801
+ 2081,
2802
+ 533,
2803
+ 583,
2804
+ 654,
2805
+ 263,
2806
+ 1936,
2807
+ 5289,
2808
+ 1940,
2809
+ 342,
2810
+ 451,
2811
+ 447,
2812
+ 5442,
2813
+ 263,
2814
+ 2713,
2815
+ 286,
2816
+ 12187,
2817
+ 63,
2818
+ 12287
2819
+ ],
2820
+ "completion_tokens": 100,
2821
+ "stopped_on_eos": false,
2822
+ "completion": " that the human race is not only a highly advanced civilization, but also a civilization founded on a shared philosophy that believes in the power of living and the dignity of the earth. The concept of humanity transcends the limits of human understanding, but the human spirit is built on a complex interplay of human intelligence and cultural heritage.\nuser\nThis is fascinating! Can you tell me more about the specific cultural values that have shaped the concept of humanity?",
2823
+ "full_text": "In a distant future, humanity discovered that the human race is not only a highly advanced civilization, but also a civilization founded on a shared philosophy that believes in the power of living and the dignity of the earth. The concept of humanity transcends the limits of human understanding, but the human spirit is built on a complex interplay of human intelligence and cultural heritage.\nuser\nThis is fascinating! Can you tell me more about the specific cultural values that have shaped the concept of humanity?",
2824
+ "full_text_sha256": "6455cdbe15a71ae11ff476739071f31b3f86b8bf4d97f3dabe99a7d1595c5c7c"
2825
+ },
2826
+ {
2827
+ "prompt_id": "code",
2828
+ "prompt": "def fibonacci(n):",
2829
+ "prompt_token_ids": [
2830
+ 1830,
2831
+ 7545,
2832
+ 262,
2833
+ 387,
2834
+ 1659,
2835
+ 40,
2836
+ 110,
2837
+ 1000
2838
+ ],
2839
+ "completion_token_ids": [
2840
+ 470,
2841
+ 982,
2842
+ 303,
2843
+ 1001,
2844
+ 7545,
2845
+ 262,
2846
+ 387,
2847
+ 1659,
2848
+ 40,
2849
+ 110,
2850
+ 41,
2851
+ 10,
2852
+ 35,
2853
+ 332,
2854
+ 573,
2855
+ 303,
2856
+ 2035,
2857
+ 346,
2858
+ 58,
2859
+ 10,
2860
+ 35,
2861
+ 445,
2862
+ 982,
2863
+ 303,
2864
+ 784,
2865
+ 7545,
2866
+ 262,
2867
+ 387,
2868
+ 1659,
2869
+ 40,
2870
+ 110,
2871
+ 486,
2872
+ 346,
2873
+ 41,
2874
+ 10,
2875
+ 35,
2876
+ 332,
2877
+ 982,
2878
+ 7545,
2879
+ 262,
2880
+ 387,
2881
+ 1659,
2882
+ 40,
2883
+ 110,
2884
+ 41,
2885
+ 10,
2886
+ 35,
2887
+ 10,
2888
+ 35,
2889
+ 332,
2890
+ 982,
2891
+ 7545,
2892
+ 262,
2893
+ 387,
2894
+ 1659,
2895
+ 40,
2896
+ 50,
2897
+ 41,
2898
+ 10,
2899
+ 35,
2900
+ 10,
2901
+ 35,
2902
+ 332,
2903
+ 982,
2904
+ 7545,
2905
+ 262,
2906
+ 387,
2907
+ 1659,
2908
+ 40,
2909
+ 52,
2910
+ 41,
2911
+ 10,
2912
+ 35,
2913
+ 10,
2914
+ 35,
2915
+ 332,
2916
+ 982,
2917
+ 7545,
2918
+ 262,
2919
+ 387,
2920
+ 1659,
2921
+ 40,
2922
+ 54,
2923
+ 41,
2924
+ 10,
2925
+ 35,
2926
+ 10,
2927
+ 35,
2928
+ 332,
2929
+ 982,
2930
+ 7545,
2931
+ 262,
2932
+ 387,
2933
+ 1659,
2934
+ 40,
2935
+ 929,
2936
+ 41,
2937
+ 10,
2938
+ 35,
2939
+ 10
2940
+ ],
2941
+ "completion_tokens": 100,
2942
+ "stopped_on_eos": false,
2943
+ "completion": "\n return n * fibonacci(n)\n# if n > 1:\n# return n + fibonacci(n - 1)\n# return fibonacci(n)\n#\n# return fibonacci(2)\n#\n# return fibonacci(4)\n#\n# return fibonacci(6)\n#\n# return fibonacci(10)\n#\n",
2944
+ "full_text": "def fibonacci(n):\n return n * fibonacci(n)\n# if n > 1:\n# return n + fibonacci(n - 1)\n# return fibonacci(n)\n#\n# return fibonacci(2)\n#\n# return fibonacci(4)\n#\n# return fibonacci(6)\n#\n# return fibonacci(10)\n#\n",
2945
+ "full_text_sha256": "ad3aa2ea2cab9a46ae5f52a9cc109bcf5f7362e1d214ceb791ee2dd6995a345f"
2946
+ }
2947
+ ]
2948
  }
2949
  ]
2950
  }
release_manifest.json CHANGED
@@ -1,19 +1,19 @@
1
  {
2
  "schema_version": 3,
3
- "revision": 5,
4
- "revision_id": "r005",
5
- "release": "Pollock 1.3",
6
  "model_id": "SlayerLab/pollock-mini-lm-125m",
7
  "publication": {
8
- "weights_commit": "e780025f15e06bc3765a74d973906eb8a11c022c",
9
- "version_tag": "v1.3",
10
- "publication_commit_pending": false
11
  },
12
  "source_checkpoint": {
13
- "path_in_training_workspace": "runs/pollock-r005-v1p3/checkpoints/ckpt-final.pt",
14
- "sha256": "bc68b2cb8a736fb00fa26b40c425dc1944b2ea02eb9514937ce5ca0908bbe9d2",
15
- "iteration": 22078,
16
- "tokens_seen": 10851778560,
17
  "native_nanogpt_parameters": 126637952,
18
  "native_unique_trainable_parameters": 127555456
19
  },
@@ -28,14 +28,17 @@
28
  "tied_word_embeddings": true
29
  },
30
  "training": {
31
- "dataset": "SlayerLab/minimal-en-corpus-5b",
32
- "training_tokens": 5425901705,
33
- "validation_tokens": 5269579,
34
- "tokenizer_sha256": "3733307577230bb4802d2c774d8e2323f7f64e4139c13712736a57daf91bdda1",
 
 
35
  "init_from": "scratch",
36
  "micro_batch_per_gpu": 12,
37
  "gradient_accumulation_global": 40,
38
- "ddp_world_size": 1,
 
39
  "effective_batch_tokens": 491520,
40
  "optimizer": "fused AdamW",
41
  "learning_rate": 0.0004,
@@ -47,26 +50,28 @@
47
  "weight_decay": 0.1,
48
  "grad_clip": 1.0,
49
  "precision": "bfloat16",
50
- "hardware": "1x NVIDIA GeForce RTX 5090",
51
  "framework": "PyTorch 2.8.0+cu128",
52
  "nanogpt_commit": "3adf61e154c3fe3fca428ad6bc3818b27a3b8291",
53
- "data_pass_equivalent": 1.9999954201160746,
54
- "runtime_hours": 16.855,
55
- "mean_tokens_per_second": 182508.3
56
  },
57
  "evaluation": {
58
  "protocol": "fixed sampled subset",
59
  "subset_tokens_per_split": 1228800,
60
- "final_train_loss": 2.447470265626907,
61
- "final_validation_loss": 2.5406661581993104,
62
- "best_validation_loss": 2.5406045150756835,
63
  "best_validation_update": 22000,
64
  "best_validation_tokens_seen": 10813440000,
65
- "final_validation_perplexity": 12.68812045164865,
66
  "benchmark_harness": "lm-evaluation-harness 0.4.12",
67
  "benchmark_num_fewshot": 0,
68
  "benchmark_batch_size": 8,
69
  "truncated_benchmark_requests": 0,
 
 
70
  "results_file": "benchmarks/english.json"
71
  },
72
  "fixed_inference": {
@@ -81,43 +86,44 @@
81
  "dtype": "float32",
82
  "seed": 1337,
83
  "prompt_count": 4,
84
- "revision_count": 5,
85
  "local_review_artifact": false,
86
  "exact_replay_verified": true,
87
- "publication_commit_pending": false
88
  },
89
  "conversion": {
90
  "target_class": "GPT2LMHeadModel",
91
  "transformers_version": "5.15.1",
92
- "checkpoint_sha256": "bc68b2cb8a736fb00fa26b40c425dc1944b2ea02eb9514937ce5ca0908bbe9d2",
93
- "model_sha256": "3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c",
94
  "unique_serialized_parameters": 127674624,
95
  "compatibility_zero_bias_parameters": 119168,
96
  "validation_probe_shape": [2, 64],
97
- "max_absolute_logit_error": 4.76837158203125e-06
98
  },
99
  "artifacts": {
100
  ".gitattributes": {"sha256": "c759491a998899dbefaec4d51cc791e68c714a37dd9f9829020a368788fe3063"},
101
- "CHANGELOG.md": {"sha256": "74a87cefef26ff31a9b9828df875d8c3a68383e13e360ff7ce201cfceb9af915"},
102
- "LICENSE.md": {"sha256": "162b97898a956d43d0a5428d1999bac3e42e2a97eb01a79f73ad719a4a15c412"},
103
- "README.md": {"sha256": "1958e268e5ac6838941b8225d27f506bbce4e37c8f76a9045d5913129e874ad7"},
104
  "assets/pollock-mini-lm-avatar-320.png": {"sha256": "7be10cc9d0f4aedeb298d9a5d722b2b2ac5b2e219b3884916dbca16198f70750"},
105
- "benchmarks/english.json": {"sha256": "c9210b71334898140b543ce684e9209c1919bbcdceb43224f4c5770f4fb93db7"},
106
- "config.json": {"sha256": "ba6775a805995b19d738604bcb035fccf012a0f4596b156d9945fd3a55c50d0b"},
107
  "generation_config.json": {"sha256": "435beb27be51f0ed054f4a011e5109d125cdadc118b8799b18b155cc798d94d2"},
108
- "inference-samples/README.md": {"sha256": "178c85ec0e003754371bde7aea0af55fe918c8d45f8cad79ec371264f52210f1"},
109
- "inference-samples/config.json": {"sha256": "fb504c65e8c6e01aa8d7f235c7f2ab0a88b95b2acbd884ec1e1d0e083669cdec"},
110
  "inference-samples/generate.py": {"sha256": "807a9946472bd0087dd2b9260ee90d1dea09785d24d4f923f028667c53a23933"},
111
  "inference-samples/requirements.txt": {"sha256": "584583335ffb3061aa62058aee725b0de25fed7523904fbd13fb4622f601943d"},
112
- "inference-samples/results.json": {"sha256": "c2eee21056ddd95eb3ece7d5bcd93242e62d90302814c0178c6beb4425d81553"},
113
- "model.safetensors": {"sha256": "3db43dfa44622e8156f060fee6f16875e108db4f5f949969cb91c0ca8cdce32c"},
114
  "special_tokens_map.json": {"sha256": "8b2257a17ea997bb038f43b133aefec82344ad2b8abc2b8a02a6c0a994ed624e"},
115
- "tokenizer.json": {"sha256": "3733307577230bb4802d2c774d8e2323f7f64e4139c13712736a57daf91bdda1"},
116
  "tokenizer_config.json": {"sha256": "4cdabe37dbdc1adfcc017ee9a1f86ab89bdf184827d2d9f05181cae0f8af19bf"},
117
- "training-history/r001.md": {"sha256": "b914562e1b67c92d279cdda55dda93965c35737475b000a758db95564a8c895e"},
118
- "training-history/r002.md": {"sha256": "2b777a46e34fad856f323742dd0adf7f2c8ba6ee73a2ba2106d727644b1c8ca8"},
119
- "training-history/r003.md": {"sha256": "9fb6a1e04bdadada6f8760cc64a96e538ee7d23b7fe112959d8ba45b89ca8ada"},
120
- "training-history/r004.md": {"sha256": "968191c3a314989e5bfc749cd79e105ae06d5622c24b0f870435e698c969ac63"},
121
- "training-history/r005.md": {"sha256": "a25c2932701695dd7402582510adbe71e50a6fcc2da0fb3c6224a846dfdb22df"}
 
122
  }
123
  }
 
1
  {
2
  "schema_version": 3,
3
+ "revision": 6,
4
+ "revision_id": "r006",
5
+ "release": "Pollock 1.4",
6
  "model_id": "SlayerLab/pollock-mini-lm-125m",
7
  "publication": {
8
+ "weights_commit": "a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9",
9
+ "version_tag": "v1.4",
10
+ "publication_commit_pending": true
11
  },
12
  "source_checkpoint": {
13
+ "path_in_training_workspace": "runs/r006-corrected-corpus-v2-s1337-lr0.0004/checkpoints/ckpt-final.pt",
14
+ "sha256": "580fbb3c039ea1f349285a9bf7a67731fb6d455de931b6b7601d4ed09fce9437",
15
+ "iteration": 22003,
16
+ "tokens_seen": 10814914560,
17
  "native_nanogpt_parameters": 126637952,
18
  "native_unique_trainable_parameters": 127555456
19
  },
 
28
  "tied_word_embeddings": true
29
  },
30
  "training": {
31
+ "dataset": "SlayerLab/minimal-en-corpus-2.5b",
32
+ "dataset_revision": "d68d992622e9fc11f19e7d7fb8547c4e653439a4",
33
+ "training_tokens": 2689323439,
34
+ "validation_tokens": 5236486,
35
+ "tokenizer_sha256": "6cda4e5ec8293f3b821e02253f9c0e88e43ad4f7c6e1324763e1fb91749fef51",
36
+ "native_tokenizer_sha256": "d0d126f0a59e51e8cb1b26d77b5527870be0e8ea1e84adefd730fff906c48234",
37
  "init_from": "scratch",
38
  "micro_batch_per_gpu": 12,
39
  "gradient_accumulation_global": 40,
40
+ "gradient_accumulation_per_gpu": 20,
41
+ "ddp_world_size": 2,
42
  "effective_batch_tokens": 491520,
43
  "optimizer": "fused AdamW",
44
  "learning_rate": 0.0004,
 
50
  "weight_decay": 0.1,
51
  "grad_clip": 1.0,
52
  "precision": "bfloat16",
53
+ "hardware": "2x NVIDIA GeForce RTX 4090",
54
  "framework": "PyTorch 2.8.0+cu128",
55
  "nanogpt_commit": "3adf61e154c3fe3fca428ad6bc3818b27a3b8291",
56
+ "data_pass_equivalent": 4.021425762020438,
57
+ "runtime_hours": 15.226956854563,
58
+ "mean_tokens_per_second": 199894.6379482684
59
  },
60
  "evaluation": {
61
  "protocol": "fixed sampled subset",
62
  "subset_tokens_per_split": 1228800,
63
+ "final_train_loss": 2.48974818944931,
64
+ "final_validation_loss": 2.536356544494629,
65
+ "best_validation_loss": 2.5362311387062073,
66
  "best_validation_update": 22000,
67
  "best_validation_tokens_seen": 10813440000,
68
+ "final_validation_perplexity": 12.633557211556399,
69
  "benchmark_harness": "lm-evaluation-harness 0.4.12",
70
  "benchmark_num_fewshot": 0,
71
  "benchmark_batch_size": 8,
72
  "truncated_benchmark_requests": 0,
73
+ "benchmark_checkpoint_sha256": "9339d20ea32d24bc9777518a2f0ec553e4a3871647d342db92d7c5b6c18c7b56",
74
+ "benchmark_tokenizer_sha256": "6cda4e5ec8293f3b821e02253f9c0e88e43ad4f7c6e1324763e1fb91749fef51",
75
  "results_file": "benchmarks/english.json"
76
  },
77
  "fixed_inference": {
 
86
  "dtype": "float32",
87
  "seed": 1337,
88
  "prompt_count": 4,
89
+ "revision_count": 6,
90
  "local_review_artifact": false,
91
  "exact_replay_verified": true,
92
+ "publication_commit_pending": true
93
  },
94
  "conversion": {
95
  "target_class": "GPT2LMHeadModel",
96
  "transformers_version": "5.15.1",
97
+ "checkpoint_sha256": "580fbb3c039ea1f349285a9bf7a67731fb6d455de931b6b7601d4ed09fce9437",
98
+ "model_sha256": "8e74aa4d34464229e86a1831fb34d442f91e22aa600e0cf2503891957fdc6a43",
99
  "unique_serialized_parameters": 127674624,
100
  "compatibility_zero_bias_parameters": 119168,
101
  "validation_probe_shape": [2, 64],
102
+ "max_absolute_logit_error": 0.0
103
  },
104
  "artifacts": {
105
  ".gitattributes": {"sha256": "c759491a998899dbefaec4d51cc791e68c714a37dd9f9829020a368788fe3063"},
106
+ "CHANGELOG.md": {"sha256": "8422ad337b22be7adb561d1b2388cf53e5a6d1c6d49d628468498c72f078769a"},
107
+ "LICENSE.md": {"sha256": "46cbe928ed0aa24875f02f774313b27ec9d8abdf41f0c9ef67e0adb4e0614de4"},
108
+ "README.md": {"sha256": "875fa9cd748691e376df905e036a0c471db77aaa1272d71c5ea10d0f76fa888d"},
109
  "assets/pollock-mini-lm-avatar-320.png": {"sha256": "7be10cc9d0f4aedeb298d9a5d722b2b2ac5b2e219b3884916dbca16198f70750"},
110
+ "benchmarks/english.json": {"sha256": "646a3aff7c97859e6974baa2c191702829a982af87675870dd1a339f8073a146"},
111
+ "config.json": {"sha256": "1529f8a8fc31b4fa68fc18fffbad7e155dc8f9e8663a40b4f3306b4012ea38c0"},
112
  "generation_config.json": {"sha256": "435beb27be51f0ed054f4a011e5109d125cdadc118b8799b18b155cc798d94d2"},
113
+ "inference-samples/README.md": {"sha256": "6559c85e1ed70f47f2d2978f28151fdd9e3355e397c5591cbe0b8ab767fa5ac5"},
114
+ "inference-samples/config.json": {"sha256": "a1b306a1e1f3a70942706d00107a83dd98289a534d17af78adbb95d48c9b69c1"},
115
  "inference-samples/generate.py": {"sha256": "807a9946472bd0087dd2b9260ee90d1dea09785d24d4f923f028667c53a23933"},
116
  "inference-samples/requirements.txt": {"sha256": "584583335ffb3061aa62058aee725b0de25fed7523904fbd13fb4622f601943d"},
117
+ "inference-samples/results.json": {"sha256": "a279806ea225130b0e0aa36d9bd13653401e236395fe64b5f0aa7c825d445620"},
118
+ "model.safetensors": {"sha256": "8e74aa4d34464229e86a1831fb34d442f91e22aa600e0cf2503891957fdc6a43"},
119
  "special_tokens_map.json": {"sha256": "8b2257a17ea997bb038f43b133aefec82344ad2b8abc2b8a02a6c0a994ed624e"},
120
+ "tokenizer.json": {"sha256": "6cda4e5ec8293f3b821e02253f9c0e88e43ad4f7c6e1324763e1fb91749fef51"},
121
  "tokenizer_config.json": {"sha256": "4cdabe37dbdc1adfcc017ee9a1f86ab89bdf184827d2d9f05181cae0f8af19bf"},
122
+ "training-history/r001.md": {"sha256": "5b0ac39265921ed3604e7dfe8181b00d07902640d8024f591462eeb508bcfece"},
123
+ "training-history/r002.md": {"sha256": "80031a84b2b67dc4cb4f81a178fa354feea15cb184d17368ee7e64ec3d0a2629"},
124
+ "training-history/r003.md": {"sha256": "7ea9e6b15a0af635ff238417b9fff2a98c12a9090800a0923d622253b7eab02d"},
125
+ "training-history/r004.md": {"sha256": "e260bfba69f154caf16f405d6944c2f963f467da9ed1e6c3c94e08932feae9b9"},
126
+ "training-history/r005.md": {"sha256": "a25c2932701695dd7402582510adbe71e50a6fcc2da0fb3c6224a846dfdb22df"},
127
+ "training-history/r006.md": {"sha256": "47d53f9aa8a233eb7a319f559209936df988e83d96f6f6517fa3670b5dd2cdd3"}
128
  }
129
  }
training-history/r001.md CHANGED
@@ -2,7 +2,7 @@
2
 
3
  ## Identity
4
 
5
- `r001` is the first published experimental predecessor of Pollock 1.0. All values and samples in this file describe the r001 weights, not the current r005 weights.
6
 
7
  | Field | Value |
8
  |---|---|
@@ -101,7 +101,7 @@ def fibonacci(n):)) = (n - 1) + (n - 1) + ( n - 1) + ( n - 2) + ( n - 1) + ( n -
101
 
102
  ## Fixed cross-revision inference reference
103
 
104
- The later [`fixed-sampling-v1`](../inference-samples/README.md) suite reruns the same four prompts on r001-r005 using immutable commits for published revisions, verified weight hashes, CPU float32, a pinned software environment, and exact output token IDs. Use that suite, rather than the historical samples above, for direct revision-to-revision comparison.
105
 
106
  ## Limitations
107
 
 
2
 
3
  ## Identity
4
 
5
+ `r001` is the first published experimental predecessor of Pollock 1.0. All values and samples in this file describe the r001 weights, not the current r006 weights.
6
 
7
  | Field | Value |
8
  |---|---|
 
101
 
102
  ## Fixed cross-revision inference reference
103
 
104
+ The later [`fixed-sampling-v1`](../inference-samples/README.md) suite reruns the same four prompts on r001-r006 using immutable commits for published revisions, verified weight hashes, CPU float32, a pinned software environment, and exact output token IDs. Use that suite, rather than the historical samples above, for direct revision-to-revision comparison.
105
 
106
  ## Limitations
107
 
training-history/r002.md CHANGED
@@ -104,7 +104,7 @@ Structured results: [`../benchmarks/english.json`](../benchmarks/english.json).
104
 
105
  ## Fixed inference reference
106
 
107
- The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r002 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r005.
108
 
109
  ## Licensing and limitations
110
 
 
104
 
105
  ## Fixed inference reference
106
 
107
+ The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r002 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r006.
108
 
109
  ## Licensing and limitations
110
 
training-history/r003.md CHANGED
@@ -123,7 +123,7 @@ Structured English results: [`../benchmarks/english.json`](../benchmarks/english
123
 
124
  ## Fixed inference reference
125
 
126
- The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r003 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r005.
127
 
128
  ## Release preparation status
129
 
 
123
 
124
  ## Fixed inference reference
125
 
126
+ The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r003 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r006.
127
 
128
  ## Release preparation status
129
 
training-history/r004.md CHANGED
@@ -132,7 +132,7 @@ Structured English results: [`../benchmarks/english.json`](../benchmarks/english
132
 
133
  ## Fixed inference reference
134
 
135
- The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r004 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r005.
136
 
137
  ## Release preparation status
138
 
 
132
 
133
  ## Fixed inference reference
134
 
135
+ The shared [`fixed-sampling-v1`](../inference-samples/README.md) suite loads r004 from its immutable published commit, verifies the weight hash above, and records the exact output token IDs for the same four prompts used across r001-r006.
136
 
137
  ## Release preparation status
138
 
training-history/r006.md CHANGED
@@ -46,4 +46,4 @@ LAMBADA perplexity was **49.907207**. Benchmarks used complete splits, zero few-
46
 
47
  ## Conversion and release status
48
 
49
- The native final checkpoint was converted to `GPT2LMHeadModel` with Transformers 5.15.1. A deterministic `[2, 64]` parity probe produced maximum absolute logit error **0.0**. Publication and immutable remote replay details are recorded in `release_manifest.json`.
 
46
 
47
  ## Conversion and release status
48
 
49
+ The native final checkpoint was converted to `GPT2LMHeadModel` with Transformers 5.15.1. A deterministic `[2, 64]` parity probe produced maximum absolute logit error **0.0**. The fixed inference suite replayed exactly against immutable Hugging Face weights commit [`a2e53d9`](https://huggingface.co/SlayerLab/pollock-mini-lm-125m/commit/a2e53d9f1b690ebfbea375cf8bd3b5bdf69dece9), with the expected model-file SHA-256 verified before loading. Merge and version tag `v1.4` remain pending.