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": 77672.48214877312,
-      "required_pages_per_second_upper_estimate": 0.005373562540040868
+      "required_token_positions_per_second": 77659.08017146889,
+      "required_pages_per_second_upper_estimate": 0.005372635360154089
     },
     {
       "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": 9709.06026859664,
-      "required_pages_per_second_upper_estimate": 0.004064980786227735
+      "required_token_positions_per_second": 9707.38502143361,
+      "required_pages_per_second_upper_estimate": 0.004064279395223713
     },
     {
       "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": 19418.12053719328,
-      "required_pages_per_second_upper_estimate": 0.004044282417622215
+      "required_token_positions_per_second": 19414.77004286722,
+      "required_pages_per_second_upper_estimate": 0.004043584598012575
     },
     {
       "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": 29127.180805789925,
-      "required_pages_per_second_upper_estimate": 0.00174613001467718
+      "required_token_positions_per_second": 29122.15506430083,
+      "required_pages_per_second_upper_estimate": 0.0017458287291487718
     },
     {
       "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": 29127.180805789925,
-      "required_pages_per_second_upper_estimate": 0.0022645101907216164
+      "required_token_positions_per_second": 29122.15506430083,
+      "required_pages_per_second_upper_estimate": 0.00226411946142674
     },
     {
       "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": 29127.180805789925,
-      "required_pages_per_second_upper_estimate": 0.37361220104654796
+      "required_token_positions_per_second": 29122.15506430083,
+      "required_pages_per_second_upper_estimate": 0.3735477362309466
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04575339884374241,
-    "producer_p95_seconds": 0.13111309700002494,
-    "gpu_update_seconds": 0.9492164787367307,
-    "producer_batches_per_second": 21.85629975633532,
-    "gpu_batches_per_second": 1.0535004631723786,
-    "producer_headroom": 20.746359892923078,
+    "producer_batch_seconds": 0.04706735642187354,
+    "producer_p95_seconds": 0.12111215400000219,
+    "gpu_update_seconds": 0.9493802893005019,
+    "producer_batches_per_second": 21.246147564286648,
+    "gpu_batches_per_second": 1.053318687221529,
+    "producer_headroom": 20.17067372110361,
     "sustainable": true,
-    "buffer_seconds": 60.74985463915076,
+    "buffer_seconds": 60.76033851523212,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 52,
+    "upload_pages_per_commit": 54,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -3,3 +3,4 @@
 {"step": 645300, "loss": 0.5487010478973389, "lr": 0.0001, "grad_norm": 0.17951284983949858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94455290557, "tokens_per_second": 154418.1227540069, "data_wait_s": 1.4860999954180443e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 645400, "loss": 0.516700029373169, "lr": 0.0001, "grad_norm": 0.16708887996363114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94470019608, "tokens_per_second": 154878.31277059382, "data_wait_s": 9.39000005928392e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 645500, "loss": 0.5504806041717529, "lr": 0.0001, "grad_norm": 0.1736480488849375, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94484547368, "tokens_per_second": 152933.4877733488, "data_wait_s": 9.080000040739833e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 645600, "loss": 0.54863041639328, "lr": 0.0001, "grad_norm": 0.18142711624629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94499089731, "tokens_per_second": 151655.98749001828, "data_wait_s": 9.263000038117752e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10ab6cd004ffa09c60c3948a9d86802711666010b0f998e6dc4571c551a991f4
-size 10755
+oid sha256:4578631aef1b4410c462b64a3a935b7a9f15197808d8e3687bc7344c3f3f9889
+size 12873