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": 77520.70324439358,
-      "required_pages_per_second_upper_estimate": 0.005363062123260312
+      "required_token_positions_per_second": 76086.20199893185,
+      "required_pages_per_second_upper_estimate": 0.005263819998597802
     },
     {
       "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": 9690.087905549197,
-      "required_pages_per_second_upper_estimate": 0.004057037454007762
+      "required_token_positions_per_second": 9510.775249866481,
+      "required_pages_per_second_upper_estimate": 0.003981962989547459
     },
     {
       "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": 19380.175811098394,
-      "required_pages_per_second_upper_estimate": 0.0040363795318463696
+      "required_token_positions_per_second": 19021.550499732963,
+      "required_pages_per_second_upper_estimate": 0.003961687337074406
     },
     {
       "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": 29070.26371664759,
-      "required_pages_per_second_upper_estimate": 0.0017427179220904602
+      "required_token_positions_per_second": 28532.325749599444,
+      "required_pages_per_second_upper_estimate": 0.00171046936234964
     },
     {
       "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": 29070.26371664759,
-      "required_pages_per_second_upper_estimate": 0.002260085137392617
+      "required_token_positions_per_second": 28532.325749599444,
+      "required_pages_per_second_upper_estimate": 0.0022182628265936793
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29070.26371664759,
-      "required_pages_per_second_upper_estimate": 0.37288212973983903
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28532.325749599444,
+      "required_pages_per_second_upper_estimate": 0.3659820390913334
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11376639882814743,
-    "producer_p95_seconds": 0.34233429299638374,
-    "gpu_update_seconds": 0.9510749633883402,
-    "producer_batches_per_second": 8.789941584690345,
-    "gpu_batches_per_second": 1.0514418300292097,
-    "producer_headroom": 8.359893370845018,
+    "producer_batch_seconds": 0.5663860027658529,
+    "producer_p95_seconds": 0.35269591699761804,
+    "gpu_update_seconds": 0.9690061806611802,
+    "producer_batches_per_second": 1.765580355299503,
+    "gpu_batches_per_second": 1.0319851616608595,
+    "producer_headroom": 1.710858276739181,
     "sustainable": true,
-    "buffer_seconds": 60.86879765685377,
+    "buffer_seconds": 60.07838320099317,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297441.18719505536,
-    "upload_bytes_per_second": 5011585.056821735,
+    "generation_bytes_per_second": 289752.69782077294,
+    "upload_bytes_per_second": 4887616.851467316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.82395469559427,
+    "estimated_full_shards_per_hour": 62.17418385474578,
     "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": 1.1389114759986114,
-      "bytes": 7332921,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8940556989982724,
       "bytes": 4992530,
@@ -562,6 +556,12 @@
       "bytes": 6231149,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.980561084994406,
+      "bytes": 658594,
+      "pages": 20,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -382,3 +382,4 @@
 {"step": 595300, "loss": 0.5223484635353088, "lr": 0.0003, "grad_norm": 0.16592439938990478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87133793712, "tokens_per_second": 152539.51282323984, "data_wait_s": 1.0310002835467458e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595400, "loss": 0.5727034211158752, "lr": 0.0003, "grad_norm": 0.15256439488777832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87148499372, "tokens_per_second": 154078.71587574232, "data_wait_s": 1.5449004422407597e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 595500, "loss": 0.5839543342590332, "lr": 0.0003, "grad_norm": 0.20130340175974581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87163181146, "tokens_per_second": 153542.6390328098, "data_wait_s": 1.160599640570581e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 595600, "loss": 0.5854461789131165, "lr": 0.0003, "grad_norm": 0.16052127449733206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87177776574, "tokens_per_second": 129848.64415712895, "data_wait_s": 1.2450000212993473e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d11027552233a564a294f0eec333eb5de64fabf97e1b3c27dbd676b54232ddd
-size 819741
+oid sha256:9f647ea1415e2f032ab75767ef177c55c4254bc01b4555d06dbd44f70dc16ab7
+size 821859