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": 76909.99438628563,
-      "required_pages_per_second_upper_estimate": 0.005320811867416624
+      "required_token_positions_per_second": 76408.61879976107,
+      "required_pages_per_second_upper_estimate": 0.005286125540989215
     },
     {
       "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": 9613.749298285704,
-      "required_pages_per_second_upper_estimate": 0.004025076073277932
+      "required_token_positions_per_second": 9551.077349970134,
+      "required_pages_per_second_upper_estimate": 0.0039988366372573055
     },
     {
       "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": 19227.498596571408,
-      "required_pages_per_second_upper_estimate": 0.004004580894429307
+      "required_token_positions_per_second": 19102.154699940267,
+      "required_pages_per_second_upper_estimate": 0.003978475066301883
     },
     {
       "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": 28841.24789485711,
-      "required_pages_per_second_upper_estimate": 0.0017289887732610334
+      "required_token_positions_per_second": 28653.232049910403,
+      "required_pages_per_second_upper_estimate": 0.0017177175104400508
     },
     {
       "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": 28841.24789485711,
-      "required_pages_per_second_upper_estimate": 0.002242280164582549
+      "required_token_positions_per_second": 28653.232049910403,
+      "required_pages_per_second_upper_estimate": 0.0022276627596322447
     },
     {
       "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": 28841.24789485711,
-      "required_pages_per_second_upper_estimate": 0.36994456067594195
+      "required_token_positions_per_second": 28653.232049910403,
+      "required_pages_per_second_upper_estimate": 0.3675328952926515
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.29055152728116695,
-    "producer_p95_seconds": 0.3067423400025291,
-    "gpu_update_seconds": 0.9586270365551732,
-    "producer_batches_per_second": 3.4417303166756343,
-    "gpu_batches_per_second": 1.0431585610119036,
-    "producer_headroom": 3.299335734096861,
+    "producer_batch_seconds": 0.12746728186724,
+    "producer_p95_seconds": 0.2951981939986581,
+    "gpu_update_seconds": 0.9649173242250852,
+    "producer_batches_per_second": 7.84515042096467,
+    "gpu_batches_per_second": 1.036358219397801,
+    "producer_headroom": 7.5699215523405305,
     "sustainable": true,
-    "buffer_seconds": 60.393503302975915,
+    "buffer_seconds": 60.78979142618037,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 161,
+    "upload_pages_per_commit": 152,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 561541.0274371954,
+    "generation_bytes_per_second": 580024.14227707,
     "upload_bytes_per_second": 5757134.689290455,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.24681624096345,
+    "estimated_full_shards_per_hour": 62.229839330835695,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 587600, "loss": 0.6157953143119812, "lr": 0.0003, "grad_norm": 0.16558787912803208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86007028597, "tokens_per_second": 152591.07657914417, "data_wait_s": 1.0386997018940747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587700, "loss": 0.6037683486938477, "lr": 0.0003, "grad_norm": 0.1436797045756794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86021631913, "tokens_per_second": 152464.6715341517, "data_wait_s": 1.5053999959491193e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 587800, "loss": 0.5790461897850037, "lr": 0.0003, "grad_norm": 0.18055201919238653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86036408163, "tokens_per_second": 153070.66596862036, "data_wait_s": 1.0163999832002446e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 587900, "loss": 0.5522806644439697, "lr": 0.0003, "grad_norm": 0.13940387312785324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86050949198, "tokens_per_second": 151149.34521239338, "data_wait_s": 1.1436997738201171e-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 sha256:482da16cead469f28c6e4d50216a557a786c04bfc78412ccc0d8ef7383a4f09d
-size 655089
+oid sha256:5f3eddfe71e2e9550fda3f61ece25e37b2d3649c83b2896919b2597505e41f81
+size 657207