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": 76927.82963734682,
-      "required_pages_per_second_upper_estimate": 0.00532204575146852
+      "required_token_positions_per_second": 77356.07424559617,
+      "required_pages_per_second_upper_estimate": 0.005351672707131602
     },
     {
       "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": 9615.978704668352,
-      "required_pages_per_second_upper_estimate": 0.004026009479175048
+      "required_token_positions_per_second": 9669.509280699522,
+      "required_pages_per_second_upper_estimate": 0.004048421613513757
     },
     {
       "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": 19231.957409336705,
-      "required_pages_per_second_upper_estimate": 0.004005509547541518
+      "required_token_positions_per_second": 19339.018561399043,
+      "required_pages_per_second_upper_estimate": 0.004027807562123659
     },
     {
       "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": 28847.936114005057,
-      "required_pages_per_second_upper_estimate": 0.0017293897217866333
+      "required_token_positions_per_second": 29008.527842098567,
+      "required_pages_per_second_upper_estimate": 0.0017390169506764687
     },
     {
       "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": 28847.936114005057,
-      "required_pages_per_second_upper_estimate": 0.002242800144202935
+      "required_token_positions_per_second": 29008.527842098567,
+      "required_pages_per_second_upper_estimate": 0.002255285444693845
     },
     {
       "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": 28847.936114005057,
-      "required_pages_per_second_upper_estimate": 0.37003034996992157
+      "required_token_positions_per_second": 29008.527842098567,
+      "required_pages_per_second_upper_estimate": 0.37209024822794173
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4764918981561266,
-    "producer_p95_seconds": 0.24925029900259688,
-    "gpu_update_seconds": 0.9584047846867453,
-    "producer_batches_per_second": 2.098671570009238,
-    "gpu_batches_per_second": 1.0434004670864097,
-    "producer_headroom": 2.011376874182897,
+    "producer_batch_seconds": 0.09266152628907776,
+    "producer_p95_seconds": 0.24963428000046406,
+    "gpu_update_seconds": 0.9530990386859928,
+    "producer_batches_per_second": 10.79196555515698,
+    "gpu_batches_per_second": 1.049208906325903,
+    "producer_headroom": 10.285811996152463,
     "sustainable": true,
-    "buffer_seconds": 60.379501435264956,
+    "buffer_seconds": 60.045239437217546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286774.6283284259,
-    "upload_bytes_per_second": 5411937.717162252,
+    "generation_bytes_per_second": 285245.4326180876,
+    "upload_bytes_per_second": 5521753.069656111,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.5351594675978,
+    "estimated_full_shards_per_hour": 61.20702966601345,
     "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": 1.6240387940015353,
-      "bytes": 5730323,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.3926623539991851,
       "bytes": 6408754,
@@ -562,6 +556,12 @@
       "bytes": 6591396,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1665754960013146,
+      "bytes": 11803743,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -252,3 +252,4 @@
 {"step": 582300, "loss": 0.5519033074378967, "lr": 0.0003, "grad_norm": 0.17627105548430186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85231989464, "tokens_per_second": 153438.6389978105, "data_wait_s": 1.1235999409109354e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582400, "loss": 0.587166965007782, "lr": 0.0003, "grad_norm": 0.16023193403620564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85246743268, "tokens_per_second": 152593.23711836973, "data_wait_s": 1.982600224437192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582500, "loss": 0.6072389483451843, "lr": 0.0003, "grad_norm": 0.1896262142150918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85261442459, "tokens_per_second": 148580.8885633686, "data_wait_s": 8.775001333560795e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 582600, "loss": 0.542369544506073, "lr": 0.0003, "grad_norm": 0.15535529138038562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85276068054, "tokens_per_second": 152024.50757107403, "data_wait_s": 9.142000635620207e-06, "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 sha256:59761138389a54847ca27a01c16610bccb65545ee2aa82542e568f2081011963
-size 542052
+oid sha256:e05e738457e5863de58daf797f56b5712a0196ed23630ca1e07df77389039853
+size 544170