Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75462.82451005068,
-      "required_pages_per_second_upper_estimate": 0.006540926369751045
+      "required_token_positions_per_second": 75818.90130844405,
+      "required_pages_per_second_upper_estimate": 0.006571790204167388
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9432.853063756334,
-      "required_pages_per_second_upper_estimate": 0.0037629061208538112
+      "required_token_positions_per_second": 9477.362663555507,
+      "required_pages_per_second_upper_estimate": 0.0037806616656915217
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18865.70612751267,
-      "required_pages_per_second_upper_estimate": 0.004364249087512277
+      "required_token_positions_per_second": 18954.725327111013,
+      "required_pages_per_second_upper_estimate": 0.0043848421126549495
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28298.559191269003,
-      "required_pages_per_second_upper_estimate": 0.0017232337836826542
+      "required_token_positions_per_second": 28432.08799066652,
+      "required_pages_per_second_upper_estimate": 0.0017313649869944957
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28298.559191269003,
-      "required_pages_per_second_upper_estimate": 0.0021785014219694197
+      "required_token_positions_per_second": 28432.08799066652,
+      "required_pages_per_second_upper_estimate": 0.0021887808385784146
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28298.559191269003,
-      "required_pages_per_second_upper_estimate": 0.3513733959704112
+      "required_token_positions_per_second": 28432.08799066652,
+      "required_pages_per_second_upper_estimate": 0.3530313767667845
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.4861114322891069,
-    "producer_p95_seconds": 0.25144685200029926,
-    "gpu_update_seconds": 0.9770108722895786,
-    "producer_batches_per_second": 2.057141497970914,
-    "gpu_batches_per_second": 1.023530063341616,
-    "producer_headroom": 2.009849609355653,
+    "producer_batch_seconds": 0.10116095686716164,
+    "producer_p95_seconds": 0.22836887300036324,
+    "gpu_update_seconds": 0.9724224266988792,
+    "producer_batches_per_second": 9.885236666089849,
+    "gpu_batches_per_second": 1.0283596640142694,
+    "producer_headroom": 9.61262582733183,
     "sustainable": true,
-    "buffer_seconds": 60.57467408195387,
+    "buffer_seconds": 60.29019045533051,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 239537.13045601017,
+    "generation_bytes_per_second": 240213.38520610376,
     "upload_bytes_per_second": 5783638.709949139,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 51.3990920568488,
+    "estimated_full_shards_per_hour": 51.544200583813996,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -132,3 +132,4 @@
 {"step": 368200, "loss": 0.5461985785514116, "lr": 0.0003, "grad_norm": 0.16204623314239167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53894327568, "tokens_per_second": 149660.57647001426, "data_wait_s": 1.8304999684914947e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368300, "loss": 0.5524621438235044, "lr": 0.0003, "grad_norm": 0.1714633989948716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53908853175, "tokens_per_second": 147898.06913565728, "data_wait_s": 1.995999991777353e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 368400, "loss": 0.5785097442567348, "lr": 0.0003, "grad_norm": 0.1597017862073903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53923384418, "tokens_per_second": 148113.3441642095, "data_wait_s": 1.0154999472433701e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 368500, "loss": 0.5769478399306536, "lr": 0.0003, "grad_norm": 0.16911262658034257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53938125446, "tokens_per_second": 150158.0008052164, "data_wait_s": 1.0830000974237919e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:31318350b984875696fe2bab8567d02fee6c44c2665d49a9aacc260baa41ed87
-size 285543
+oid sha256:a5a1d35bb9c26811a666dbc654a11625181a7fffacbf78c1b05ac4200607d342
+size 287661