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": 76416.61308240582,
-      "required_pages_per_second_upper_estimate": 0.00528667860401187
+      "required_token_positions_per_second": 76852.54571105684,
+      "required_pages_per_second_upper_estimate": 0.005316837434765004
     },
     {
       "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": 9552.076635300728,
-      "required_pages_per_second_upper_estimate": 0.0039992550171577745
+      "required_token_positions_per_second": 9606.568213882105,
+      "required_pages_per_second_upper_estimate": 0.00402206950319624
     },
     {
       "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": 19104.153270601455,
-      "required_pages_per_second_upper_estimate": 0.003978891315864755
+      "required_token_positions_per_second": 19213.13642776421,
+      "required_pages_per_second_upper_estimate": 0.004001589633422629
     },
     {
       "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": 28656.229905902183,
-      "required_pages_per_second_upper_estimate": 0.0017178972273292968
+      "required_token_positions_per_second": 28819.704641646313,
+      "required_pages_per_second_upper_estimate": 0.001727697288125689
     },
     {
       "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": 28656.229905902183,
-      "required_pages_per_second_upper_estimate": 0.0022278958297494308
+      "required_token_positions_per_second": 28819.704641646313,
+      "required_pages_per_second_upper_estimate": 0.0022406052714041657
     },
     {
       "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": 28656.229905902183,
-      "required_pages_per_second_upper_estimate": 0.3675713485704671
+      "required_token_positions_per_second": 28819.704641646313,
+      "required_pages_per_second_upper_estimate": 0.36966822695509693
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08658311354689374,
-    "producer_p95_seconds": 0.16852038499928312,
-    "gpu_update_seconds": 0.9648163799211241,
-    "producer_batches_per_second": 11.549596209177627,
-    "gpu_batches_per_second": 1.036466648795652,
-    "producer_headroom": 11.143239604089496,
+    "producer_batch_seconds": 0.0481229498750082,
+    "producer_p95_seconds": 0.16451592700013862,
+    "gpu_update_seconds": 0.9593436276944655,
+    "producer_batches_per_second": 20.780106011733338,
+    "gpu_batches_per_second": 1.0423793634854714,
+    "producer_headroom": 19.93526228517183,
     "sustainable": true,
-    "buffer_seconds": 60.78343193503082,
+    "buffer_seconds": 60.438648544751324,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411944.9665028221,
+    "generation_bytes_per_second": 410638.11289468105,
     "upload_bytes_per_second": 4965264.387060935,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.39380022347925,
+    "estimated_full_shards_per_hour": 88.11337986116718,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 689900, "loss": 0.5641610622406006, "lr": 8e-05, "grad_norm": 0.19524551050481168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100982806168, "tokens_per_second": 150606.61389982823, "data_wait_s": 1.5345999599958304e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690000, "loss": 0.5735451579093933, "lr": 8e-05, "grad_norm": 0.20603365382569494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100997233939, "tokens_per_second": 150022.19957227245, "data_wait_s": 1.6033000065363012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690100, "loss": 0.5196791887283325, "lr": 8e-05, "grad_norm": 0.19404623849519337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101011784041, "tokens_per_second": 141949.50861257964, "data_wait_s": 1.574199995957315e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 690200, "loss": 0.5197702050209045, "lr": 8e-05, "grad_norm": 0.19167683174711828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101026325514, "tokens_per_second": 151782.48313986367, "data_wait_s": 1.764999979059212e-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:c450e0dae2bbc21f7dbafb5635fef6ddc59f7d781efa98f2f6d595f1f0768291
-size 108966
+oid sha256:6e3cd17dc32c54f3382d3b3414a89f64fa7d96a63f019ba74d4ef592e6309d99
+size 111084