Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77422.81057901727,
-      "required_pages_per_second_upper_estimate": 0.005356289681527302
+      "required_token_positions_per_second": 77326.47236719842,
+      "required_pages_per_second_upper_estimate": 0.005349624780498225
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9677.851322377159,
-      "required_pages_per_second_upper_estimate": 0.0040519142521625445
+      "required_token_positions_per_second": 9665.809045899803,
+      "required_pages_per_second_upper_estimate": 0.004046872402472804
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19355.702644754318,
-      "required_pages_per_second_upper_estimate": 0.004031282416697696
+      "required_token_positions_per_second": 19331.618091799606,
+      "required_pages_per_second_upper_estimate": 0.004026266239469604
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29033.55396713148,
-      "required_pages_per_second_upper_estimate": 0.0017405172286595115
+      "required_token_positions_per_second": 28997.42713769941,
+      "required_pages_per_second_upper_estimate": 0.0017383514803975333
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29033.55396713148,
-      "required_pages_per_second_upper_estimate": 0.002257231115836863
+      "required_token_positions_per_second": 28997.42713769941,
+      "required_pages_per_second_upper_estimate": 0.002254422413753649
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29033.55396713148,
-      "required_pages_per_second_upper_estimate": 0.37241125648890444
+      "required_token_positions_per_second": 28997.42713769941,
+      "required_pages_per_second_upper_estimate": 0.37194786031091714
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08861391272660057,
-    "producer_p95_seconds": 0.2398417510012223,
-    "gpu_update_seconds": 0.9522774935269707,
-    "producer_batches_per_second": 11.284909662947486,
-    "gpu_batches_per_second": 1.0501140757787717,
-    "producer_headroom": 10.746365488509923,
+    "producer_batch_seconds": 0.339315747538933,
+    "producer_p95_seconds": 0.24388737499975832,
+    "gpu_update_seconds": 0.9534639010802091,
+    "producer_batches_per_second": 2.9471075458567104,
+    "gpu_batches_per_second": 1.0488074051540586,
+    "producer_headroom": 2.80996065757546,
     "sustainable": true,
-    "buffer_seconds": 60.94575958572612,
+    "buffer_seconds": 60.06822576805317,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352184.4028317846,
-    "upload_bytes_per_second": 5577688.163350661,
+    "generation_bytes_per_second": 344332.62921411166,
+    "upload_bytes_per_second": 5471685.619773158,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.57057441439775,
+    "estimated_full_shards_per_hour": 73.8857665759803,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.575513838999541,
-      "bytes": 18103820,
-      "pages": 39,
-      "shards": 5
-    },
     {
       "seconds": 0.8145781550010724,
       "bytes": 930729,
@@ -562,6 +556,12 @@
       "bytes": 6065248,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.0634998090026784,
+      "bytes": 7147043,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -240,3 +240,4 @@
 {"step": 581100, "loss": 0.5525113940238953, "lr": 0.0003, "grad_norm": 0.1515899625324291, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85056007800, "tokens_per_second": 152747.6060758848, "data_wait_s": 9.957999282050878e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 581200, "loss": 0.553583025932312, "lr": 0.0003, "grad_norm": 0.18004174666705045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85070744118, "tokens_per_second": 153300.87512376314, "data_wait_s": 9.076997230295092e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 581300, "loss": 0.5507333278656006, "lr": 0.0003, "grad_norm": 0.16153292466350272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85085361082, "tokens_per_second": 152954.14175127208, "data_wait_s": 1.0825999197550118e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 581400, "loss": 0.5653789043426514, "lr": 0.0003, "grad_norm": 0.1666213727706197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85100087281, "tokens_per_second": 153145.8002052712, "data_wait_s": 1.8106999050360173e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:773e35d7fa59202fdcff524418800315a506fa583d5304190e21df5797d2bce2
-size 516636
+oid sha256:c4e324c19ff934d85b4ece11864371c8de9ad1e2536e7e42d3f4fdce55f1139b
+size 518754