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": 76913.31628393629,
-      "required_pages_per_second_upper_estimate": 0.0053210416840039606
+      "required_token_positions_per_second": 76252.78488579558,
+      "required_pages_per_second_upper_estimate": 0.005275344589236586
     },
     {
       "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": 9614.164535492037,
-      "required_pages_per_second_upper_estimate": 0.004025249924425085
+      "required_token_positions_per_second": 9531.598110724448,
+      "required_pages_per_second_upper_estimate": 0.003990681086557909
     },
     {
       "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": 19228.329070984073,
-      "required_pages_per_second_upper_estimate": 0.004004753860348382
+      "required_token_positions_per_second": 19063.196221448896,
+      "required_pages_per_second_upper_estimate": 0.003970361042636298
     },
     {
       "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": 28842.493606476106,
-      "required_pages_per_second_upper_estimate": 0.0017290634517705007
+      "required_token_positions_per_second": 28594.794332173344,
+      "required_pages_per_second_upper_estimate": 0.0017142142584909426
     },
     {
       "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": 28842.493606476106,
-      "required_pages_per_second_upper_estimate": 0.0022423770131816197
+      "required_token_positions_per_second": 28594.794332173344,
+      "required_pages_per_second_upper_estimate": 0.002223119484118544
     },
     {
       "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": 28842.493606476106,
-      "required_pages_per_second_upper_estimate": 0.3699605393270495
+      "required_token_positions_per_second": 28594.794332173344,
+      "required_pages_per_second_upper_estimate": 0.36678331899505323
     }
   ],
   "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": 0.10738558146056221,
-    "producer_p95_seconds": 0.3126368610028294,
-    "gpu_update_seconds": 0.9585856333098778,
-    "producer_batches_per_second": 9.312237140209126,
-    "gpu_batches_per_second": 1.0432036171323824,
-    "producer_headroom": 8.92657673657913,
+    "producer_batch_seconds": 0.46141952385136165,
+    "producer_p95_seconds": 0.30270002099859994,
+    "gpu_update_seconds": 0.9668892763775516,
+    "producer_batches_per_second": 2.1672251569531173,
+    "gpu_batches_per_second": 1.0342445866671492,
+    "producer_headroom": 2.0954667637536257,
     "sustainable": true,
-    "buffer_seconds": 60.3908948985223,
+    "buffer_seconds": 60.91402441178575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 38893.31844869009,
-    "upload_bytes_per_second": 5158491.84100879,
+    "generation_bytes_per_second": 39195.78314134223,
+    "upload_bytes_per_second": 5224038.148757857,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.34560074897315,
+    "estimated_full_shards_per_hour": 8.410502630998614,
     "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": 0.8940000050024537,
-      "bytes": 393500,
-      "pages": 10,
-      "shards": 1
-    },
     {
       "seconds": 1.0701383749983506,
       "bytes": 5409485,
@@ -562,6 +556,12 @@
       "bytes": 2568385,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 0.9927107309995336,
+      "bytes": 6005616,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 717900, "loss": 0.551626443862915, "lr": 8e-05, "grad_norm": 0.2137054670401659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105080967918, "tokens_per_second": 151987.31471702774, "data_wait_s": 1.6012003470677882e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718000, "loss": 0.5673214793205261, "lr": 8e-05, "grad_norm": 0.2272528752780872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105095720485, "tokens_per_second": 152871.38338677492, "data_wait_s": 1.561100361868739e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718100, "loss": 0.5424805879592896, "lr": 8e-05, "grad_norm": 0.21388578976032796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105110379791, "tokens_per_second": 135666.51715236792, "data_wait_s": 1.4131997886579484e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 718200, "loss": 0.4762612581253052, "lr": 8e-05, "grad_norm": 0.22901733781678954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105124954836, "tokens_per_second": 151718.2090392901, "data_wait_s": 2.0215004042256624e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5179749511439b89627fc96e3c5f902edeb1248cdf3f2e8d61c7424ff4391fc
-size 705921
+oid sha256:5074141d7674fa3cd6f2ca394d28bf0ae5656a66102e724009a0c1a6790d2a18
+size 708039