Maggio33 commited on
Commit
6484a05
·
verified ·
1 Parent(s): b2c38fd

Model card: archive complete 18:16 UTC, training paused (8ae72199)

Browse files
Files changed (3) hide show
  1. EVALUATION.json +1 -1
  2. README.md +3 -3
  3. TRAINING_CONFIGURATION.md +1 -1
EVALUATION.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "schema_version": 1,
3
  "measurement_owner": "SlayerLab / Fabryka AI",
4
- "status": "descriptive intermediate base checkpoint measurements; at the 12 November 2026 14:53 UTC archive snapshot the separate repository was public and native .pt upload was in progress; verify individual file availability and checksums in that repository; main repository contains documentation; lineage checks incomplete",
5
  "source_aggregate_sha256": "d91456c7480f8031288afd5056c002ce68ef7d7751556d222ae54d7ab6fc648a",
6
  "source_generator_sha256": null,
7
  "benchmark_community_report": "https://github.com/slayerlabs/benchmarks-fabryka-ai/blob/8b547007526971bd2fc35dfe56ff2c7d9b802568/research/results/full-run-2026-11/README.md",
 
1
  {
2
  "schema_version": 1,
3
  "measurement_owner": "SlayerLab / Fabryka AI",
4
+ "status": "descriptive intermediate base checkpoint measurements; snapshot 12 November 2026 18:51 UTC: archive public, all eight native files uploaded at 18:16 UTC, milestones through update 35000; operators report training paused since approximately 16:48 UTC at around update 40000 while allocation prices exceed approved ceilings; reported progress is not a saved-state or export receipt; 70% EN / 30% PL approved at 18:44 UTC for future c16+ segments, not confirmed runtime adoption or cumulative composition; verify archive SHA256SUMS and file sizes; main repository contains documentation; lineage checks incomplete",
5
  "source_aggregate_sha256": "d91456c7480f8031288afd5056c002ce68ef7d7751556d222ae54d7ab6fc648a",
6
  "source_generator_sha256": null,
7
  "benchmark_community_report": "https://github.com/slayerlabs/benchmarks-fabryka-ai/blob/8b547007526971bd2fc35dfe56ff2c7d9b802568/research/results/full-run-2026-11/README.md",
README.md CHANGED
@@ -31,14 +31,14 @@ tags:
31
  | 30,000 | 31,457,280,000 | 50,205,910,554 | `wagi_30000.pt` |
32
  | 35,000 | 36,700,160,000 | 55,448,790,554 | `wagi_35000.pt` |
33
 
34
- Files are managed in the separate [checkpoint archive](https://huggingface.co/SlayerLab/Slayer1B-1.09-checkpoints). **Archive snapshot, 12 November 2026, 14:53 UTC:** the repository is public following the owner's decision at 14:42 UTC. Upload of the eight native checkpoint files started at 14:50 UTC and is in progress. A file becomes available only when it appears in the archive's file list; verify its checksum and size before use. This main repository contains documentation during training.
35
 
36
  The archive distinguishes seven weights-only `.pt` files from the full update-35,000 recovery checkpoint `ckpt_35000.pt`, which also contains optimizer and queue state. These are native training checkpoints, not a Transformers `AutoModel` or safetensors release. Check the archive's exact revision, file list and checksums before use; its publication status is separate from the evaluation table below.
37
 
38
  Token counts: **positions in this continuation = updates × 1,048,576**, counting each logical update once, not repeated compute after recovery. **Lineage lower bound = 18,863,257,682 inherited positions + positions in this continuation − 114,627,128 identified-copy positions**. The lineage figure is a source- and file-disjoint lower bound, **not globally unique tokens**. Exact evaluated safetensors identities in [EVALUATION.json](EVALUATION.json) are distinct from the native `.pt` file digests in the archive.
39
 
40
 
41
- **Execution receipt reported by the operators, 12 November 2026, 15:44 UTC:** training resumed at approximately **15:41 UTC** from update **38,200** on an eight-GPU B200 spot instance. Weights-only milestones through update **35,000** and a full update-35,000 recovery checkpoint are retained on the trusted workstation. The [CPT Track history](https://track.fabryka.ai/run/71cf98ec-7c60-54a8-ac47-0ea7a23f573a) provides later telemetry; a published point and a saved recovery checkpoint have separate receipts.
42
 
43
 
44
  ## Objective and starting point
@@ -62,7 +62,7 @@ The [training configuration](TRAINING_CONFIGURATION.md) records the 120,000-upda
62
  | Setting | Operator-selected recipe |
63
  |---|---|
64
  | Selected scheduling argument | `KROKI=120000`; confirmed runtime `--steps 120000` adoption at 11 November, 16:23:47 UTC, resuming from update 29,800 on H200; configured decay begins at update 96,000 |
65
- | New-source allocation | **20% PL / 80% EN target**; not a measured cumulative model-language ratio |
66
  | Optimizer / schedule | Muon peak 0.02 plus AdamW peak 0.0006; 2,000-update warmup, then constant learning rates under the main-base plan; adoption receipt confirms LR 0.0006. The configured final-20% WSD decay to 0.00006 remains at update 96,000; an extension approaching it requires review |
67
  | Checkpoints | Runtime `--ckpt-every 200` confirmed at the 16:23:47 UTC adoption; weights-only milestones every 5,000 updates. Native milestone files and full-state recovery files are distinguished in the separate checkpoint archive |
68
  | Code | Launcher v5 SHA256 `796c71016c1e167bfd3ca98b78518f91c3502a96eac233dd7186ef6639fcc656`; exact command and converter pins in the configuration document |
 
31
  | 30,000 | 31,457,280,000 | 50,205,910,554 | `wagi_30000.pt` |
32
  | 35,000 | 36,700,160,000 | 55,448,790,554 | `wagi_35000.pt` |
33
 
34
+ Files are managed in the separate [checkpoint archive](https://huggingface.co/SlayerLab/Slayer1B-1.09-checkpoints). **Archive snapshot, 12 November 2026, 18:51 UTC:** upload of all eight native checkpoint files completed at **18:16 UTC**. The archive is public; verify its exact revision, file list, `SHA256SUMS` and file sizes before use. This main repository contains documentation during training.
35
 
36
  The archive distinguishes seven weights-only `.pt` files from the full update-35,000 recovery checkpoint `ckpt_35000.pt`, which also contains optimizer and queue state. These are native training checkpoints, not a Transformers `AutoModel` or safetensors release. Check the archive's exact revision, file list and checksums before use; its publication status is separate from the evaluation table below.
37
 
38
  Token counts: **positions in this continuation = updates × 1,048,576**, counting each logical update once, not repeated compute after recovery. **Lineage lower bound = 18,863,257,682 inherited positions + positions in this continuation − 114,627,128 identified-copy positions**. The lineage figure is a source- and file-disjoint lower bound, **not globally unique tokens**. Exact evaluated safetensors identities in [EVALUATION.json](EVALUATION.json) are distinct from the native `.pt` file digests in the archive.
39
 
40
 
41
+ **Execution receipt reported by the operators, 12 November 2026, 18:51 UTC:** training has been paused since approximately **16:48 UTC**, at reported progress around **update 40,000**, while allocation prices exceed the approved ceilings. The public native archive currently ends at update **35,000**. The [CPT Track history](https://track.fabryka.ai/run/71cf98ec-7c60-54a8-ac47-0ea7a23f573a) provides later telemetry; reported progress, saved recovery state and published files have separate receipts.
42
 
43
 
44
  ## Objective and starting point
 
62
  | Setting | Operator-selected recipe |
63
  |---|---|
64
  | Selected scheduling argument | `KROKI=120000`; confirmed runtime `--steps 120000` adoption at 11 November, 16:23:47 UTC, resuming from update 29,800 on H200; configured decay begins at update 96,000 |
65
+ | New-source allocation | Initially **20% PL / 80% EN target**. On 12 November at **18:44 UTC**, the owner approved **30% PL / 70% EN** for future, not-yet-built segments from **c16** onward; this is not a runtime-adoption receipt or a measured cumulative model-language ratio |
66
  | Optimizer / schedule | Muon peak 0.02 plus AdamW peak 0.0006; 2,000-update warmup, then constant learning rates under the main-base plan; adoption receipt confirms LR 0.0006. The configured final-20% WSD decay to 0.00006 remains at update 96,000; an extension approaching it requires review |
67
  | Checkpoints | Runtime `--ckpt-every 200` confirmed at the 16:23:47 UTC adoption; weights-only milestones every 5,000 updates. Native milestone files and full-state recovery files are distinguished in the separate checkpoint archive |
68
  | Code | Launcher v5 SHA256 `796c71016c1e167bfd3ca98b78518f91c3502a96eac233dd7186ef6639fcc656`; exact command and converter pins in the configuration document |
TRAINING_CONFIGURATION.md CHANGED
@@ -59,7 +59,7 @@ Inherited-count footprint: counter SHA256 `abb39b8ce019c1a18a19ccf0ef4c32977c0c5
59
 
60
  Operators confirmed the 04:50 UTC start: fresh 1.07 hardware control within the 0.001 tolerance, pinned 1.08 weights/config converted to the training format, step-zero `INIT-FROM`, fresh optimizer, world size eight, c01/queue/CL validation identities and `PRE-FLIGHT OK`. The inherited 1.08 checkpoint step 1,709 is source metadata, **not the new continuation's initial training counter**.
61
 
62
- The distinct [CPT Track run](https://track.fabryka.ai/run/71cf98ec-7c60-54a8-ac47-0ea7a23f573a) preserves the experiment's published updates. Its 02:27 UTC registration and the reported 04:50 UTC training start are separate events. **Operator receipt, 12 November 2026, 15:44 UTC:** training resumed at approximately **15:41 UTC** from update **38,200** on an eight-GPU B200 spot instance. The trusted workstation retained weights-only milestones through **35,000** and a full update-35,000 recovery checkpoint. **Archive snapshot, 14:53 UTC:** the separate archive is public and upload of its eight native files is in progress. The [model card](README.md) and the archive's actual file list describe availability.
63
 
64
 
65
  ## Historical code identities
 
59
 
60
  Operators confirmed the 04:50 UTC start: fresh 1.07 hardware control within the 0.001 tolerance, pinned 1.08 weights/config converted to the training format, step-zero `INIT-FROM`, fresh optimizer, world size eight, c01/queue/CL validation identities and `PRE-FLIGHT OK`. The inherited 1.08 checkpoint step 1,709 is source metadata, **not the new continuation's initial training counter**.
61
 
62
+ The distinct [CPT Track run](https://track.fabryka.ai/run/71cf98ec-7c60-54a8-ac47-0ea7a23f573a) preserves the experiment's published updates. Its 02:27 UTC registration and the reported 04:50 UTC training start are separate events. **Operator receipt, 12 November 2026, 18:51 UTC:** training has been paused since approximately **16:48 UTC**, at reported progress around **update 40,000**, while allocation prices exceed the approved ceilings. **Archive snapshot, 18:51 UTC:** upload of all eight native checkpoint files completed at **18:16 UTC**; the separate public archive contains milestones through update **35,000**. On **12 November, 18:44 UTC**, the owner approved **70% EN / 30% PL** for future, not-yet-built segments from **c16** onward; approval is not a runtime-adoption receipt or a measured cumulative language ratio. Reported progress, saved recovery state and published files have separate receipts. The [model card](README.md) and the archive's actual file list describe availability.
63
 
64
 
65
  ## Historical code identities