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": 77109.08324740753,
-      "required_pages_per_second_upper_estimate": 0.005334585296778839
+      "required_token_positions_per_second": 76245.46945138989,
+      "required_pages_per_second_upper_estimate": 0.0052748384905102535
     },
     {
       "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": 9638.635405925941,
-      "required_pages_per_second_upper_estimate": 0.004035495366865877
+      "required_token_positions_per_second": 9530.683681423736,
+      "required_pages_per_second_upper_estimate": 0.003990298234105155
     },
     {
       "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": 19277.270811851882,
-      "required_pages_per_second_upper_estimate": 0.004014947134290645
+      "required_token_positions_per_second": 19061.36736284747,
+      "required_pages_per_second_upper_estimate": 0.003969980139619866
     },
     {
       "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": 28915.906217777825,
-      "required_pages_per_second_upper_estimate": 0.0017334644257234718
+      "required_token_positions_per_second": 28592.051044271208,
+      "required_pages_per_second_upper_estimate": 0.0017140498025699676
     },
     {
       "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": 28915.906217777825,
-      "required_pages_per_second_upper_estimate": 0.002248084520802377
+      "required_token_positions_per_second": 28592.051044271208,
+      "required_pages_per_second_upper_estimate": 0.002222906205550601
     },
     {
       "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": 28915.906217777825,
-      "required_pages_per_second_upper_estimate": 0.37090219748050723
+      "required_token_positions_per_second": 28592.051044271208,
+      "required_pages_per_second_upper_estimate": 0.3667481310433577
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.05964572272665691,
-    "producer_p95_seconds": 0.2018800480000209,
-    "gpu_update_seconds": 0.9561519459833391,
-    "producer_batches_per_second": 16.765661547648232,
-    "gpu_batches_per_second": 1.0458588765110612,
-    "producer_headroom": 16.030519914481896,
+    "producer_batch_seconds": 0.11466135003908562,
+    "producer_p95_seconds": 0.20930288700037636,
+    "gpu_update_seconds": 0.9669820453660543,
+    "producer_batches_per_second": 8.721334605419536,
+    "gpu_batches_per_second": 1.0341453647378185,
+    "producer_headroom": 8.433373975070332,
     "sustainable": true,
-    "buffer_seconds": 60.237572596950365,
+    "buffer_seconds": 60.91986885806142,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386950.5865983246,
-    "upload_bytes_per_second": 3652605.4285908067,
+    "generation_bytes_per_second": 385092.3134157955,
+    "upload_bytes_per_second": 3586563.771010249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.03058813535979,
+    "estimated_full_shards_per_hour": 82.63184596877478,
     "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": 3.149820212999657,
-      "bytes": 17435942,
-      "pages": 43,
-      "shards": 5
-    },
     {
       "seconds": 2.0315385989997594,
       "bytes": 5979371,
@@ -562,6 +556,12 @@
       "bytes": 10666543,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.843333258999337,
+      "bytes": 1195587,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -159,3 +159,4 @@
 {"step": 940900, "loss": 0.5127492547035217, "lr": 3e-05, "grad_norm": 0.33479694834108586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137720947183, "tokens_per_second": 150850.5953767437, "data_wait_s": 1.6580999727011658e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941000, "loss": 0.5244556069374084, "lr": 3e-05, "grad_norm": 0.336522169660098, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137735627709, "tokens_per_second": 153399.61329272506, "data_wait_s": 1.2283999240025878e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 941100, "loss": 0.5392598509788513, "lr": 3e-05, "grad_norm": 0.36996917852931727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137750354229, "tokens_per_second": 153107.729645144, "data_wait_s": 1.302700002270285e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 941200, "loss": 0.5398966670036316, "lr": 3e-05, "grad_norm": 0.3430767914103548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137765008400, "tokens_per_second": 153301.6113073163, "data_wait_s": 1.3201999536249787e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:626fa47c40227965fbe25a2a0cd22e1231efb873d09da8c4fca5ececf62b4dc9
-size 343512
+oid sha256:533dfa381b64ee8d26c8f331b60c590c60b137d3ec26bddaba2eea30839ad27e
+size 345630