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": 76759.81490505325,
-      "required_pages_per_second_upper_estimate": 0.0053104221024405625
+      "required_token_positions_per_second": 77601.44694193365,
+      "required_pages_per_second_upper_estimate": 0.005368648159607325
     },
     {
       "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": 9594.976863131657,
-      "required_pages_per_second_upper_estimate": 0.0040172164466919565
+      "required_token_positions_per_second": 9700.180867741707,
+      "required_pages_per_second_upper_estimate": 0.004061263166512749
     },
     {
       "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": 19189.953726263313,
-      "required_pages_per_second_upper_estimate": 0.003996761288068943
+      "required_token_positions_per_second": 19400.361735483413,
+      "required_pages_per_second_upper_estimate": 0.004040583727557145
     },
     {
       "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": 28784.93058939497,
-      "required_pages_per_second_upper_estimate": 0.0017256126367901244
+      "required_token_positions_per_second": 29100.54260322512,
+      "required_pages_per_second_upper_estimate": 0.0017445330951076003
     },
     {
       "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": 28784.93058939497,
-      "required_pages_per_second_upper_estimate": 0.0022379017417965153
+      "required_token_positions_per_second": 29100.54260322512,
+      "required_pages_per_second_upper_estimate": 0.002262439187641273
     },
     {
       "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": 28784.93058939497,
-      "required_pages_per_second_upper_estimate": 0.369222182750285
+      "required_token_positions_per_second": 29100.54260322512,
+      "required_pages_per_second_upper_estimate": 0.37327051478592016
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08789945624997131,
+    "producer_batch_seconds": 0.3259191273593274,
     "producer_p95_seconds": 0.2462164889984706,
-    "gpu_update_seconds": 0.9605025766567649,
-    "producer_batches_per_second": 11.376634653531505,
-    "gpu_batches_per_second": 1.0411216214335566,
-    "producer_headroom": 10.927286898399652,
+    "gpu_update_seconds": 0.9500853773405538,
+    "producer_batches_per_second": 3.06824581945292,
+    "gpu_batches_per_second": 1.0525369865171124,
+    "producer_headroom": 2.915095487148504,
     "sustainable": true,
-    "buffer_seconds": 60.511662329376186,
+    "buffer_seconds": 60.80546414979544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493600.3739618888,
-    "upload_bytes_per_second": 5414231.320306781,
+    "generation_bytes_per_second": 474754.63426331966,
+    "upload_bytes_per_second": 5491742.388941373,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.91514982359406,
+    "estimated_full_shards_per_hour": 101.87129279064838,
     "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": 1.0592705330000172,
-      "bytes": 5143840,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9495372719993611,
       "bytes": 5544847,
@@ -562,6 +556,12 @@
       "bytes": 6292434,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3688695919991005,
+      "bytes": 12940523,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -227,3 +227,4 @@
 {"step": 547700, "loss": 0.5748477578163147, "lr": 0.0003, "grad_norm": 0.15021868288194715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80166016474, "tokens_per_second": 152257.38127189185, "data_wait_s": 8.578001143177971e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 547800, "loss": 0.6012737154960632, "lr": 0.0003, "grad_norm": 0.1534133025377039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80180623332, "tokens_per_second": 152703.6732536238, "data_wait_s": 6.722002581227571e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 547900, "loss": 0.5492777824401855, "lr": 0.0003, "grad_norm": 0.16366448976048606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80195120841, "tokens_per_second": 151511.10119357446, "data_wait_s": 1.6111000149976462e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 548000, "loss": 0.5742793083190918, "lr": 0.0003, "grad_norm": 0.16188139865888318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80209825350, "tokens_per_second": 153958.7221827193, "data_wait_s": 8.487000741297379e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26a660624468bd2b5d61381f7ecb2f647633d1b60e6a725f08c63226a98a2a00
-size 488319
+oid sha256:55cb5b1508f7a10839fee8b87133ab09c4b923eeeca5cee74c4cd42f698e87f5
+size 490437