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": 76632.24214134253,
-      "required_pages_per_second_upper_estimate": 0.005301596322637461
+      "required_token_positions_per_second": 76778.56218109709,
+      "required_pages_per_second_upper_estimate": 0.005311719082496949
     },
     {
       "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": 9579.030267667817,
-      "required_pages_per_second_upper_estimate": 0.004010539940174027
+      "required_token_positions_per_second": 9597.320272637136,
+      "required_pages_per_second_upper_estimate": 0.004018197583315945
     },
     {
       "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": 19158.060535335633,
-      "required_pages_per_second_upper_estimate": 0.003990118777478713
+      "required_token_positions_per_second": 19194.640545274273,
+      "required_pages_per_second_upper_estimate": 0.0039977374288691974
     },
     {
       "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": 28737.09080300345,
-      "required_pages_per_second_upper_estimate": 0.0017227447146430771
+      "required_token_positions_per_second": 28791.96081791141,
+      "required_pages_per_second_upper_estimate": 0.0017260340882551475
     },
     {
       "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": 28737.09080300345,
-      "required_pages_per_second_upper_estimate": 0.0022341824088295565
+      "required_token_positions_per_second": 28791.96081791141,
+      "required_pages_per_second_upper_estimate": 0.002238448311140961
     },
     {
       "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": 28737.09080300345,
-      "required_pages_per_second_upper_estimate": 0.3686085453368152
+      "required_token_positions_per_second": 28791.96081791141,
+      "required_pages_per_second_upper_estimate": 0.3693123589732227
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23216465640626893,
-    "producer_p95_seconds": 0.23005201699925237,
-    "gpu_update_seconds": 0.9621015637779999,
-    "producer_batches_per_second": 4.307287833898726,
-    "gpu_batches_per_second": 1.0393913050854837,
-    "producer_headroom": 4.144048360635918,
+    "producer_batch_seconds": 0.06224172532824923,
+    "producer_p95_seconds": 0.20019182600117347,
+    "gpu_update_seconds": 0.9602680475586173,
+    "producer_batches_per_second": 16.066392676716767,
+    "gpu_batches_per_second": 1.0413758976385783,
+    "producer_headroom": 15.428043526980877,
     "sustainable": true,
-    "buffer_seconds": 60.612398518014,
+    "buffer_seconds": 60.49688699619289,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342645.9241928237,
+    "generation_bytes_per_second": 343151.7799271885,
     "upload_bytes_per_second": 4000433.435844654,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.52383894289525,
+    "estimated_full_shards_per_hour": 73.63238380777112,
     "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": [

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -120,3 +120,4 @@
 {"step": 812000, "loss": 0.5742809176445007, "lr": 3e-05, "grad_norm": 0.3177930254943342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118853246833, "tokens_per_second": 149470.66895386908, "data_wait_s": 1.7865000700112432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812100, "loss": 0.535494327545166, "lr": 3e-05, "grad_norm": 0.27763589498637725, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118867908223, "tokens_per_second": 151764.48540176268, "data_wait_s": 2.0464998669922352e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 812200, "loss": 0.5729753375053406, "lr": 3e-05, "grad_norm": 0.29566897251418217, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118882688269, "tokens_per_second": 153201.9355423492, "data_wait_s": 1.7569000192452222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 812300, "loss": 0.5137185454368591, "lr": 3e-05, "grad_norm": 0.2792861971277968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118897314241, "tokens_per_second": 151032.37952066955, "data_wait_s": 1.7754000509739853e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd3a0428e074c59ae237888b2894cc8e190e1b90c929804325178af364936aa5
-size 260127
+oid sha256:e07519909baa3912d6467d61e39142cc1d0967d4247283c65c7a0f849699c248
+size 262245