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": 75662.73621161851,
-      "required_pages_per_second_upper_estimate": 0.0065582542093713695
+      "required_token_positions_per_second": 75234.38622413992,
+      "required_pages_per_second_upper_estimate": 0.006521125918099843
     },
     {
       "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": 9457.842026452314,
-      "required_pages_per_second_upper_estimate": 0.0037728745916915246
+      "required_token_positions_per_second": 9404.29827801749,
+      "required_pages_per_second_upper_estimate": 0.003751515189890494
     },
     {
       "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": 18915.684052904628,
-      "required_pages_per_second_upper_estimate": 0.0043758106009702605
+      "required_token_positions_per_second": 18808.59655603498,
+      "required_pages_per_second_upper_estimate": 0.004351037793245043
     },
     {
       "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": 28373.526079356943,
-      "required_pages_per_second_upper_estimate": 0.0017277988738463453
+      "required_token_positions_per_second": 28212.894834052466,
+      "required_pages_per_second_upper_estimate": 0.0017180172737743145
     },
     {
       "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": 28373.526079356943,
-      "required_pages_per_second_upper_estimate": 0.0021842725805360573
+      "required_token_positions_per_second": 28212.894834052466,
+      "required_pages_per_second_upper_estimate": 0.0021719067426167727
     },
     {
       "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": 28373.526079356943,
-      "required_pages_per_second_upper_estimate": 0.35230423382242876
+      "required_token_positions_per_second": 28212.894834052466,
+      "required_pages_per_second_upper_estimate": 0.3503097313539425
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11661069829682447,
-    "producer_p95_seconds": 0.3967194249999011,
-    "gpu_update_seconds": 0.9744294707211313,
-    "producer_batches_per_second": 8.575542506868187,
-    "gpu_batches_per_second": 1.026241539328593,
-    "producer_headroom": 8.356261346114131,
+    "producer_batch_seconds": 0.11569309724217192,
+    "producer_p95_seconds": 0.3877795340013108,
+    "gpu_update_seconds": 0.9799774238916225,
+    "producer_batches_per_second": 8.643558032738747,
+    "gpu_batches_per_second": 1.0204316707918282,
+    "producer_headroom": 8.470491734181056,
     "sustainable": true,
-    "buffer_seconds": 60.41462718471014,
+    "buffer_seconds": 60.75860028128059,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 411198.37627952034,
-    "upload_bytes_per_second": 5676283.9716359945,
+    "generation_bytes_per_second": 395817.8630968483,
+    "upload_bytes_per_second": 5828417.7348650275,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 88.23359934128959,
+    "estimated_full_shards_per_hour": 84.93329925231063,
     "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": 1.1431241999962367,
-      "bytes": 977191,
-      "pages": 29,
-      "shards": 1
-    },
     {
       "seconds": 0.9953306989991688,
       "bytes": 5946973,
@@ -562,6 +556,12 @@
       "bytes": 10392878,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.3803016649981146,
+      "bytes": 13473573,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 313700, "loss": 0.6515887826681137, "lr": 0.0003, "grad_norm": 0.1977031417502658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45916832172, "tokens_per_second": 150912.25244320382, "data_wait_s": 1.0141993698198348e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 313800, "loss": 0.564821919426322, "lr": 0.0003, "grad_norm": 0.152956664917729, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45931584558, "tokens_per_second": 151120.3533412845, "data_wait_s": 1.4492994523607194e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 313900, "loss": 0.6181909441947937, "lr": 0.0003, "grad_norm": 0.15279418255568192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45946273386, "tokens_per_second": 150840.54126490458, "data_wait_s": 9.335002687294036e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 314000, "loss": 0.5206610346212983, "lr": 0.0003, "grad_norm": 0.15408984745557255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45961037178, "tokens_per_second": 151764.0794378983, "data_wait_s": 1.0324001777917147e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:021807fd9b8380d230dd75685b323b1ecc5e1f324b38a580ecdbb8a34f703c09
-size 718629
+oid sha256:42294a22268f5ceffcf8231690b91dbdcaed2351aa22bd80e757b3555845d36b
+size 720747