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": 76945.67343466582,
-      "required_pages_per_second_upper_estimate": 0.0053232802267703635
+      "required_token_positions_per_second": 76604.92958791414,
+      "required_pages_per_second_upper_estimate": 0.005299706776817429
     },
     {
       "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": 9618.209179333227,
-      "required_pages_per_second_upper_estimate": 0.0040269433323396235
+      "required_token_positions_per_second": 9575.616198489268,
+      "required_pages_per_second_upper_estimate": 0.00400911054070284
     },
     {
       "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": 19236.418358666455,
-      "required_pages_per_second_upper_estimate": 0.00400643864564376
+      "required_token_positions_per_second": 19151.232396978536,
+      "required_pages_per_second_upper_estimate": 0.0039886966563290474
     },
     {
       "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": 28854.62753799968,
-      "required_pages_per_second_upper_estimate": 0.0017297908624378982
+      "required_token_positions_per_second": 28726.848595467804,
+      "required_pages_per_second_upper_estimate": 0.0017221307099402603
     },
     {
       "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": 28854.62753799968,
-      "required_pages_per_second_upper_estimate": 0.002243320372986053
+      "required_token_positions_per_second": 28726.848595467804,
+      "required_pages_per_second_upper_estimate": 0.0022333861222443695
     },
     {
       "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": 28854.62753799968,
-      "required_pages_per_second_upper_estimate": 0.3701161803722333
+      "required_token_positions_per_second": 28726.848595467804,
+      "required_pages_per_second_upper_estimate": 0.36847716929577357
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08872445987509536,
-    "producer_p95_seconds": 0.24315070399825345,
-    "gpu_update_seconds": 0.9581825294258042,
-    "producer_batches_per_second": 11.2708491143004,
-    "gpu_batches_per_second": 1.043642489076956,
-    "producer_headroom": 10.799530713116942,
+    "producer_batch_seconds": 0.1233400809999523,
+    "producer_p95_seconds": 0.23678289199961,
+    "gpu_update_seconds": 0.962444589357497,
+    "producer_batches_per_second": 8.10766453121096,
+    "gpu_batches_per_second": 1.0390208548707973,
+    "producer_headroom": 7.803177860389675,
     "sustainable": true,
-    "buffer_seconds": 60.36549935382566,
+    "buffer_seconds": 60.63400912952231,
     "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": 219142.8392799856,
-    "upload_bytes_per_second": 5119723.378959696,
+    "generation_bytes_per_second": 219401.70565923833,
+    "upload_bytes_per_second": 5009810.917321263,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 47.02295192527462,
+    "estimated_full_shards_per_hour": 47.07849862416136,
     "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": 1.3169437420001486,
-      "bytes": 11722554,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2027081380001619,
       "bytes": 2601682,
@@ -562,6 +556,12 @@
       "bytes": 10401632,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.9978380930006097,
+      "bytes": 912994,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -188,3 +188,4 @@
 {"step": 575900, "loss": 0.5540058612823486, "lr": 0.0003, "grad_norm": 0.1559572572497929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84294224818, "tokens_per_second": 151778.02242146767, "data_wait_s": 1.5068999346112832e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 576000, "loss": 0.6009294986724854, "lr": 0.0003, "grad_norm": 0.15569686658394694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84309105577, "tokens_per_second": 154206.54879942816, "data_wait_s": 1.1015999916708097e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 576100, "loss": 0.5879713892936707, "lr": 0.0003, "grad_norm": 0.19614164407461182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84323799522, "tokens_per_second": 152831.03853750153, "data_wait_s": 1.0924002708634362e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 576200, "loss": 0.616330087184906, "lr": 0.0003, "grad_norm": 0.16241952339423624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84338476900, "tokens_per_second": 152404.30469681544, "data_wait_s": 1.3199001841712743e-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 sha256:8bf37d1f32969300785c5a966917c02521e3e2901db4a4f29465e83211a14590
-size 405717
+oid sha256:2dfd751496c94ca583e1ef3e02d02cd3b5736c78f300b55a18d6fb808897d137
+size 407835