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": 77548.71689394301,
-      "required_pages_per_second_upper_estimate": 0.005365000172536764
+      "required_token_positions_per_second": 77421.13571987867,
+      "required_pages_per_second_upper_estimate": 0.005356173810885885
     },
     {
       "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": 9693.589611742877,
-      "required_pages_per_second_upper_estimate": 0.004058503545267116
+      "required_token_positions_per_second": 9677.641964984834,
+      "required_pages_per_second_upper_estimate": 0.004051826598594257
     },
     {
       "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": 19387.179223485753,
-      "required_pages_per_second_upper_estimate": 0.004037838157954251
+      "required_token_positions_per_second": 19355.283929969668,
+      "required_pages_per_second_upper_estimate": 0.004031195209450305
     },
     {
       "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": 29080.76883522863,
-      "required_pages_per_second_upper_estimate": 0.0017433476879090064
+      "required_token_positions_per_second": 29032.9258949545,
+      "required_pages_per_second_upper_estimate": 0.001740479576691512
     },
     {
       "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": 29080.76883522863,
-      "required_pages_per_second_upper_estimate": 0.00226090186415516
+      "required_token_positions_per_second": 29032.9258949545,
+      "required_pages_per_second_upper_estimate": 0.0022571822859877005
     },
     {
       "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": 29080.76883522863,
-      "required_pages_per_second_upper_estimate": 0.3730168781214791
+      "required_token_positions_per_second": 29032.9258949545,
+      "required_pages_per_second_upper_estimate": 0.3724032002533895
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4502592875782341,
-    "producer_p95_seconds": 0.22850390899839113,
-    "gpu_update_seconds": 0.9507313976687932,
-    "producer_batches_per_second": 2.2209425270905636,
-    "gpu_batches_per_second": 1.0518217894686281,
-    "producer_headroom": 2.111519792922873,
-    "sustainable": true,
-    "buffer_seconds": 60.846809450802766,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 1.4730633848125478,
+    "producer_p95_seconds": 0.22210815799917327,
+    "gpu_update_seconds": 0.9522980942408157,
+    "producer_batches_per_second": 0.6788574139511677,
+    "gpu_batches_per_second": 1.0500913590478336,
+    "producer_headroom": 0.6464746215669455,
+    "sustainable": false,
+    "buffer_seconds": 60.947078031412204,
+    "buffer_drain_seconds": 172.3980278348064,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298338.4820591867,
-    "upload_bytes_per_second": 5092014.01384785,
+    "generation_bytes_per_second": 212579.45776364848,
+    "upload_bytes_per_second": 5147180.443778019,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.01649328548146,
+    "estimated_full_shards_per_hour": 45.61460303957072,
     "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": 0.9635194839993346,
-      "bytes": 1029009,
-      "pages": 33,
-      "shards": 1
-    },
-    {
-      "seconds": 1.0835146409990557,
-      "bytes": 5946250,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 0.9391229799985013,
       "bytes": 2632765,
@@ -562,6 +550,18 @@
       "bytes": 5902816,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9907326490028936,
+      "bytes": 186236,
+      "pages": 7,
+      "shards": 1
+    },
+    {
+      "seconds": 1.1790409999994154,
+      "bytes": 12018438,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -184,3 +184,4 @@
 {"step": 575500, "loss": 0.561769962310791, "lr": 0.0003, "grad_norm": 0.15479443685832425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84235675989, "tokens_per_second": 151186.59097593135, "data_wait_s": 1.3323999155545607e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575600, "loss": 0.5846127271652222, "lr": 0.0003, "grad_norm": 0.1487987002422767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84250301136, "tokens_per_second": 131601.4857546011, "data_wait_s": 1.150900061475113e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575700, "loss": 0.5894057154655457, "lr": 0.0003, "grad_norm": 0.15861235980839236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84265001282, "tokens_per_second": 67196.43554143806, "data_wait_s": 7.2690017987042665e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 575800, "loss": 0.546024739742279, "lr": 0.0003, "grad_norm": 0.18850383590792041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84279627659, "tokens_per_second": 152615.38952956314, "data_wait_s": 1.4261000615078956e-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 sha256:03365e008789e6e85cbec250e337049bd55a712edd912daeaf92251675856e1b
-size 397245
+oid sha256:01fef13af8d4e36707c1f6b4f294c6ddddee9ea1181bf8d6f453bb4dc04ca59d
+size 399363