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": 77435.7584107616,
-      "required_pages_per_second_upper_estimate": 0.005357185442570482
+      "required_token_positions_per_second": 77419.70027780312,
+      "required_pages_per_second_upper_estimate": 0.005356074503672413
     },
     {
       "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": 9679.4698013452,
-      "required_pages_per_second_upper_estimate": 0.004052591875508779
+      "required_token_positions_per_second": 9677.46253472539,
+      "required_pages_per_second_upper_estimate": 0.004051751474891558
     },
     {
       "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": 19358.9396026904,
-      "required_pages_per_second_upper_estimate": 0.004031956589671461
+      "required_token_positions_per_second": 19354.92506945078,
+      "required_pages_per_second_upper_estimate": 0.004031120468268009
     },
     {
       "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": 29038.409404035596,
-      "required_pages_per_second_upper_estimate": 0.0017408083046880902
+      "required_token_positions_per_second": 29032.387604176165,
+      "required_pages_per_second_upper_estimate": 0.0017404473069812733
     },
     {
       "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": 29038.409404035596,
-      "required_pages_per_second_upper_estimate": 0.0022576086046993482
+      "required_token_positions_per_second": 29032.387604176165,
+      "required_pages_per_second_upper_estimate": 0.002257140436247376
     },
     {
       "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": 29038.409404035596,
-      "required_pages_per_second_upper_estimate": 0.3724735368201485
+      "required_token_positions_per_second": 29032.387604176165,
+      "required_pages_per_second_upper_estimate": 0.3723962956372566
     }
   ],
   "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": 0.1313207685232669,
-    "producer_p95_seconds": 0.3081663520024449,
-    "gpu_update_seconds": 0.9521182656842642,
-    "producer_batches_per_second": 7.614941728145795,
-    "gpu_batches_per_second": 1.0502896919862412,
-    "producer_headroom": 7.250325111488908,
+    "producer_batch_seconds": 0.08011522348422773,
+    "producer_p95_seconds": 0.2757304449987714,
+    "gpu_update_seconds": 0.9523157508417589,
+    "producer_batches_per_second": 12.48202222386448,
+    "gpu_batches_per_second": 1.0500718896186403,
+    "producer_headroom": 11.886826366143024,
     "sustainable": true,
-    "buffer_seconds": 60.93556900379291,
+    "buffer_seconds": 60.94820805387257,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425021.911897896,
+    "generation_bytes_per_second": 407918.9104417153,
     "upload_bytes_per_second": 5366755.264876879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.19980828955326,
+    "estimated_full_shards_per_hour": 87.52990231455416,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -287,3 +287,4 @@
 {"step": 673700, "loss": 0.5673876404762268, "lr": 0.0001, "grad_norm": 0.2134560078421933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98611932000, "tokens_per_second": 150182.8946675715, "data_wait_s": 1.2245000107213855e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673800, "loss": 0.5613048076629639, "lr": 0.0001, "grad_norm": 0.21319297970065137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98626528191, "tokens_per_second": 152388.1773700583, "data_wait_s": 1.0379997547715902e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 673900, "loss": 0.5678796768188477, "lr": 0.0001, "grad_norm": 0.19500271852505002, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98641213613, "tokens_per_second": 152612.64458449662, "data_wait_s": 1.095100014936179e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 674000, "loss": 0.5891708731651306, "lr": 0.0001, "grad_norm": 0.20241650769049185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98655862187, "tokens_per_second": 152752.71169859282, "data_wait_s": 1.2886001059086993e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0f113145000f12dae56821ab849412118bc3c59d779e11063ab25b082d883772
-size 616182
+oid sha256:64ceaface1c55961ea5a2bf5f0666e0955a9259ad5a14757de8a52f451045384
+size 618300