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": 76208.92207789175,
-      "required_pages_per_second_upper_estimate": 0.006605596215964511
+      "required_token_positions_per_second": 73790.54717677334,
+      "required_pages_per_second_upper_estimate": 0.006395977608850718
     },
     {
       "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": 9526.115259736469,
-      "required_pages_per_second_upper_estimate": 0.003800109805224377
+      "required_token_positions_per_second": 9223.818397096667,
+      "required_pages_per_second_upper_estimate": 0.0036795190669764908
     },
     {
       "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": 19052.230519472938,
-      "required_pages_per_second_upper_estimate": 0.004407398222875105
+      "required_token_positions_per_second": 18447.636794193335,
+      "required_pages_per_second_upper_estimate": 0.004267536105017816
     },
     {
       "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": 28578.34577920941,
-      "required_pages_per_second_upper_estimate": 0.0017402713189614454
+      "required_token_positions_per_second": 27671.455191290002,
+      "required_pages_per_second_upper_estimate": 0.001685046440244344
     },
     {
       "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": 28578.34577920941,
-      "required_pages_per_second_upper_estimate": 0.0022000401680079174
+      "required_token_positions_per_second": 27671.455191290002,
+      "required_pages_per_second_upper_estimate": 0.0021302252201160594
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28578.34577920941,
-      "required_pages_per_second_upper_estimate": 0.35484740900715706
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27671.455191290002,
+      "required_pages_per_second_upper_estimate": 0.3435868630727492
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09010654942207452,
-    "producer_p95_seconds": 0.2756729489992722,
-    "gpu_update_seconds": 0.9674457791785055,
-    "producer_batches_per_second": 11.0979724161429,
-    "gpu_batches_per_second": 1.0336496592595996,
-    "producer_headroom": 10.736686571436929,
+    "producer_batch_seconds": 0.5300723228830861,
+    "producer_p95_seconds": 0.30588025099859806,
+    "gpu_update_seconds": 0.9991523687089147,
+    "producer_batches_per_second": 1.886535019525179,
+    "gpu_batches_per_second": 1.0008483503794128,
+    "producer_headroom": 1.8849359334109013,
     "sustainable": true,
-    "buffer_seconds": 60.94908408824585,
+    "buffer_seconds": 60.94829449124379,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 224430.674793005,
-    "upload_bytes_per_second": 2213305.395430478,
+    "generation_bytes_per_second": 212404.11684350882,
+    "upload_bytes_per_second": 2214979.0557735157,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.1575983318578,
+    "estimated_full_shards_per_hour": 45.57697895983647,
     "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": 1.1981934089999413,
-      "bytes": 7536231,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.27145491400006,
       "bytes": 5090460,
@@ -562,6 +556,12 @@
       "bytes": 5469696,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.178703661000327,
+      "bytes": 7809750,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -168,3 +168,4 @@
 {"step": 297000, "loss": 0.5576042011380196, "lr": 0.0003, "grad_norm": 0.16236165534024324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43473003958, "tokens_per_second": 125207.86663961658, "data_wait_s": 1.2679000064963475e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 297100, "loss": 0.5655781496316195, "lr": 0.0003, "grad_norm": 0.14514667480996218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43487624228, "tokens_per_second": 149063.06452224927, "data_wait_s": 1.4109002222539857e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 297200, "loss": 0.6275224015116692, "lr": 0.0003, "grad_norm": 0.1596334546668923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43502440674, "tokens_per_second": 151447.01684310051, "data_wait_s": 1.048500053002499e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 297300, "loss": 0.6055130567401648, "lr": 0.0003, "grad_norm": 0.16041599723372443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43517132742, "tokens_per_second": 150014.6329969246, "data_wait_s": 1.1656000424409285e-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:beff62c6d88aeeaf9e32960902055277ec889b6ded3ab03e132e04c0fde952e2
-size 362574
+oid sha256:ec2cf4e9e21e65b0082caac82f5d2d8b875933e08d1f87141257ac166e421f78
+size 364692