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": 77250.14368994458,
-      "required_pages_per_second_upper_estimate": 0.005344344185498872
+      "required_token_positions_per_second": 77279.48846171392,
+      "required_pages_per_second_upper_estimate": 0.005346374324899127
     },
     {
       "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": 9656.267961243073,
-      "required_pages_per_second_upper_estimate": 0.004042877749567535
+      "required_token_positions_per_second": 9659.93605771424,
+      "required_pages_per_second_upper_estimate": 0.004044413504961448
     },
     {
       "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": 19312.535922486146,
-      "required_pages_per_second_upper_estimate": 0.004022291926832255
+      "required_token_positions_per_second": 19319.87211542848,
+      "required_pages_per_second_upper_estimate": 0.004023819862353725
     },
     {
       "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": 28968.80388372922,
-      "required_pages_per_second_upper_estimate": 0.0017366355600271985
+      "required_token_positions_per_second": 28979.80817314272,
+      "required_pages_per_second_upper_estimate": 0.0017372952503749608
     },
     {
       "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": 28968.80388372922,
-      "required_pages_per_second_upper_estimate": 0.002252197081657855
+      "required_token_positions_per_second": 28979.80817314272,
+      "required_pages_per_second_upper_estimate": 0.0022530526167570007
     },
     {
       "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": 28968.80388372922,
-      "required_pages_per_second_upper_estimate": 0.3715807119422432
+      "required_token_positions_per_second": 28979.80817314272,
+      "required_pages_per_second_upper_estimate": 0.37172186315135414
     }
   ],
   "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": 0.24551550085175222,
-    "producer_p95_seconds": 0.3191802799992729,
-    "gpu_update_seconds": 0.9544059917340575,
-    "producer_batches_per_second": 4.073062582732088,
-    "gpu_batches_per_second": 1.0477721312112709,
-    "producer_headroom": 3.8873553336673,
+    "producer_batch_seconds": 0.11368157182852201,
+    "producer_p95_seconds": 0.319234073998814,
+    "gpu_update_seconds": 0.9540435821663932,
+    "producer_batches_per_second": 8.796500469824663,
+    "gpu_batches_per_second": 1.0481701451512848,
+    "producer_headroom": 8.392244818759881,
     "sustainable": true,
-    "buffer_seconds": 60.12757747924562,
+    "buffer_seconds": 60.10474567648277,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303587.06248242554,
-    "upload_bytes_per_second": 4043750.9800569164,
+    "generation_bytes_per_second": 310054.4150992362,
+    "upload_bytes_per_second": 4039322.67732142,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.14271646360945,
+    "estimated_full_shards_per_hour": 66.53045978291334,
     "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": 1.275760786000319,
-      "bytes": 7467840,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.235418092001055,
       "bytes": 6152144,
@@ -562,6 +556,12 @@
       "bytes": 4898067,
       "pages": 41,
       "shards": 2
+    },
+    {
+      "seconds": 1.4413858800035086,
+      "bytes": 7671053,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -346,3 +346,4 @@
 {"step": 877100, "loss": 0.7569558620452881, "lr": 3e-05, "grad_norm": 0.7409805443240339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128383211391, "tokens_per_second": 150776.6346752798, "data_wait_s": 1.1326999810989946e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 877200, "loss": 0.5050011873245239, "lr": 3e-05, "grad_norm": 0.3247661351031221, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128397863950, "tokens_per_second": 152102.0454064988, "data_wait_s": 7.590999302919954e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 877300, "loss": 0.5222134590148926, "lr": 3e-05, "grad_norm": 0.2997342212671748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128412512155, "tokens_per_second": 152827.05755084864, "data_wait_s": 1.974500628421083e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 877400, "loss": 0.5608643293380737, "lr": 3e-05, "grad_norm": 0.32331489554311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128427011320, "tokens_per_second": 150429.5205340763, "data_wait_s": 8.13999940874055e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:168e4bb9152948c80cfec7cdc02df23b710b92823103b3c5f43bcdebc4914295
-size 742710
+oid sha256:cbc64805b1c0edb301343bcd8fabeac0d09896736cfe3a18f33c18d73e8439fe
+size 744828