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": 77039.89722667898,
-      "required_pages_per_second_upper_estimate": 0.0053297988473311525
+      "required_token_positions_per_second": 76495.34595902613,
+      "required_pages_per_second_upper_estimate": 0.005292125527101906
     },
     {
       "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": 9629.987153334872,
-      "required_pages_per_second_upper_estimate": 0.004031874524102048
+      "required_token_positions_per_second": 9561.918244878267,
+      "required_pages_per_second_upper_estimate": 0.004003375493571712
     },
     {
       "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": 19259.974306669745,
-      "required_pages_per_second_upper_estimate": 0.004011344728400722
+      "required_token_positions_per_second": 19123.836489756533,
+      "required_pages_per_second_upper_estimate": 0.003982990811333362
     },
     {
       "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": 28889.96146000462,
-      "required_pages_per_second_upper_estimate": 0.0017319090771103192
+      "required_token_positions_per_second": 28685.754734634804,
+      "required_pages_per_second_upper_estimate": 0.0017196671957300145
     },
     {
       "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": 28889.96146000462,
-      "required_pages_per_second_upper_estimate": 0.002246067430004439
+      "required_token_positions_per_second": 28685.754734634804,
+      "required_pages_per_second_upper_estimate": 0.002230191255317395
     },
     {
       "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": 28889.96146000462,
-      "required_pages_per_second_upper_estimate": 0.3705694059851031
+      "required_token_positions_per_second": 28685.754734634804,
+      "required_pages_per_second_upper_estimate": 0.3679500613721579
     }
   ],
   "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": 0.11955376807804896,
-    "producer_p95_seconds": 0.3476413300013519,
-    "gpu_update_seconds": 0.9570106224709232,
-    "producer_batches_per_second": 8.364437324528026,
-    "gpu_batches_per_second": 1.0449204810476207,
-    "producer_headroom": 8.00485537056559,
+    "producer_batch_seconds": 0.14906877539129937,
+    "producer_p95_seconds": 0.37010872200335143,
+    "gpu_update_seconds": 0.9638233421349786,
+    "producer_batches_per_second": 6.708312974162707,
+    "gpu_batches_per_second": 1.0375345317793259,
+    "producer_headroom": 6.465628630844939,
     "sustainable": true,
-    "buffer_seconds": 60.29166921566816,
+    "buffer_seconds": 60.72087055450365,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269468.2375113495,
-    "upload_bytes_per_second": 5375206.355905758,
+    "generation_bytes_per_second": 260297.0253147993,
+    "upload_bytes_per_second": 5360638.839016082,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.82161087041248,
+    "estimated_full_shards_per_hour": 55.85368222792611,
     "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": 1.2837438990027294,
-      "bytes": 5816020,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.0380413320017396,
       "bytes": 232525,
@@ -562,6 +556,12 @@
       "bytes": 11633071,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.5646861489949515,
+      "bytes": 6165459,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 482600, "loss": 0.5773614645004272, "lr": 0.0003, "grad_norm": 0.1470273699476545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70636343050, "tokens_per_second": 152910.0876266976, "data_wait_s": 1.0592004400677979e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 482700, "loss": 0.6278800368309021, "lr": 0.0003, "grad_norm": 0.1686408496044801, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70650955949, "tokens_per_second": 150788.3485357573, "data_wait_s": 7.309005013667047e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 482800, "loss": 0.6001877188682556, "lr": 0.0003, "grad_norm": 0.16184486147186183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70665634891, "tokens_per_second": 151296.96019529446, "data_wait_s": 1.2812997738365084e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 482900, "loss": 0.5762990117073059, "lr": 0.0003, "grad_norm": 0.1542558287123174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70680327522, "tokens_per_second": 152236.4964049248, "data_wait_s": 1.4051998732611537e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbb16df88177eccac89e33513eb90b73243cec2f4720dd2073a65f9d2d8d9962
-size 908697
+oid sha256:a651596a7e280cfd2beb721d2423c9e874ba1aecb39c18b09c8822870360e4c8
+size 910815