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": 77552.51462597697,
-      "required_pages_per_second_upper_estimate": 0.005365262908450826
+      "required_token_positions_per_second": 77541.0401984386,
+      "required_pages_per_second_upper_estimate": 0.0053644690809293814
     },
     {
       "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": 9694.064328247121,
-      "required_pages_per_second_upper_estimate": 0.004058702299154235
+      "required_token_positions_per_second": 9692.630024804825,
+      "required_pages_per_second_upper_estimate": 0.004058101786254607
     },
     {
       "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": 19388.128656494242,
-      "required_pages_per_second_upper_estimate": 0.0040380358998116875
+      "required_token_positions_per_second": 19385.26004960965,
+      "required_pages_per_second_upper_estimate": 0.0040374384446478775
     },
     {
       "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": 29082.19298474136,
-      "required_pages_per_second_upper_estimate": 0.001743433063497744
+      "required_token_positions_per_second": 29077.890074414478,
+      "required_pages_per_second_upper_estimate": 0.001743175110593811
     },
     {
       "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": 29082.19298474136,
-      "required_pages_per_second_upper_estimate": 0.0022610125855156028
+      "required_token_positions_per_second": 29077.890074414478,
+      "required_pages_per_second_upper_estimate": 0.00226067805316419
     },
     {
       "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": 29082.19298474136,
-      "required_pages_per_second_upper_estimate": 0.3730351455822958
+      "required_token_positions_per_second": 29077.890074414478,
+      "required_pages_per_second_upper_estimate": 0.37297995246872767
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08887171928907378,
-    "producer_p95_seconds": 0.19801124799960235,
-    "gpu_update_seconds": 0.950684840531323,
-    "producer_batches_per_second": 11.252173447295327,
-    "gpu_batches_per_second": 1.051873299505981,
-    "producer_headroom": 10.697270719372746,
+    "producer_batch_seconds": 0.0878040361953083,
+    "producer_p95_seconds": 0.2179386659990996,
+    "gpu_update_seconds": 0.9508255217020498,
+    "producer_batches_per_second": 11.38899808404746,
+    "gpu_batches_per_second": 1.0517176676220514,
+    "producer_headroom": 10.828950044928071,
     "sustainable": true,
-    "buffer_seconds": 60.84382979400467,
+    "buffer_seconds": 60.85283338893119,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 416509.9344410618,
+    "generation_bytes_per_second": 416200.06842019875,
     "upload_bytes_per_second": 3592998.9290321777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.37333607601062,
+    "estimated_full_shards_per_hour": 89.3068460412452,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 938500, "loss": 0.5211808681488037, "lr": 3e-05, "grad_norm": 0.34861600819338995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137369683862, "tokens_per_second": 154429.45637528106, "data_wait_s": 1.4294999346020631e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938600, "loss": 0.48861345648765564, "lr": 3e-05, "grad_norm": 0.32073224374380005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137384408236, "tokens_per_second": 154256.38199451554, "data_wait_s": 1.1725998774636537e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938700, "loss": 0.48941338062286377, "lr": 3e-05, "grad_norm": 0.34972946347485506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137399138972, "tokens_per_second": 153898.15498313733, "data_wait_s": 8.736999006941915e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 938800, "loss": 0.5446861386299133, "lr": 3e-05, "grad_norm": 0.35378563013557457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137413915310, "tokens_per_second": 153914.0110141069, "data_wait_s": 1.622700074221939e-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:df371c02a30240e9cc01ef3b1e2a88e2d2285def59b4eeff85ff7fc0b1537fd3
-size 291897
+oid sha256:770cc5c222f18359fd0d1b4a6a16a975b881a5cb25ed3085e065e8afd83af831
+size 294015