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": 75359.45980195286,
-      "required_pages_per_second_upper_estimate": 0.005213542287136576
+      "required_token_positions_per_second": 75756.23144108093,
+      "required_pages_per_second_upper_estimate": 0.005240991869768497
     },
     {
       "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": 9419.932475244108,
-      "required_pages_per_second_upper_estimate": 0.003943929016825921
+      "required_token_positions_per_second": 9469.528930135117,
+      "required_pages_per_second_upper_estimate": 0.003964694016797036
     },
     {
       "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": 18839.864950488216,
-      "required_pages_per_second_upper_estimate": 0.003923847028536864
+      "required_token_positions_per_second": 18939.057860270234,
+      "required_pages_per_second_upper_estimate": 0.0039445062957515166
     },
     {
       "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": 28259.797425732326,
-      "required_pages_per_second_upper_estimate": 0.0016941317054604622
+      "required_token_positions_per_second": 28408.586790405352,
+      "required_pages_per_second_upper_estimate": 0.0017030514006844048
     },
     {
       "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": 28259.797425732326,
-      "required_pages_per_second_upper_estimate": 0.0021970749481384226
+      "required_token_positions_per_second": 28408.586790405352,
+      "required_pages_per_second_upper_estimate": 0.0022086426667865
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28259.797425732326,
-      "required_pages_per_second_upper_estimate": 0.3624863383708819
+      "required_token_positions_per_second": 28408.586790405352,
+      "required_pages_per_second_upper_estimate": 0.36439484858333465
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.481568575320523,
-    "producer_p95_seconds": 0.3585090999986278,
-    "gpu_update_seconds": 0.9783509620923452,
-    "producer_batches_per_second": 2.0765474560594384,
-    "gpu_batches_per_second": 1.0221280897617304,
-    "producer_headroom": 2.031592201466163,
+    "producer_batch_seconds": 0.11900041863310662,
+    "producer_p95_seconds": 0.3881533629974001,
+    "gpu_update_seconds": 0.973226869889134,
+    "producer_batches_per_second": 8.403331782244623,
+    "gpu_batches_per_second": 1.0275096495372307,
+    "producer_headroom": 8.178348287073812,
     "sustainable": true,
-    "buffer_seconds": 60.6577596497254,
+    "buffer_seconds": 60.340065933126304,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274523.09835128055,
+    "generation_bytes_per_second": 286542.3759552319,
     "upload_bytes_per_second": 4425533.958861365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.906266335523725,
+    "estimated_full_shards_per_hour": 61.4853235148689,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -421,3 +421,4 @@
 {"step": 842100, "loss": 0.5551508665084839, "lr": 3e-05, "grad_norm": 0.31770559703554857, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123259666338, "tokens_per_second": 152730.96904519544, "data_wait_s": 1.0275995009578764e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842200, "loss": 0.47381365299224854, "lr": 3e-05, "grad_norm": 0.29832004624987274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123274331554, "tokens_per_second": 151585.23609351562, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 842300, "loss": 0.5562163591384888, "lr": 3e-05, "grad_norm": 0.2983377293134837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123289015698, "tokens_per_second": 151895.5803168626, "data_wait_s": 7.576003554277122e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 842400, "loss": 0.5182906985282898, "lr": 3e-05, "grad_norm": 0.2880123845292042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123303506551, "tokens_per_second": 150064.32580190475, "data_wait_s": 1.2801996490452439e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57a0df28df59db0c026c707fd470489e303c5c29cdf48000d1eec198338016cd
-size 902343
+oid sha256:2159469e9d0b0485e6b3e009cc5c979cc1179b511cf1f53104ef77923c179660
+size 904461