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": 77173.6762567645,
-      "required_pages_per_second_upper_estimate": 0.005339053991042585
+      "required_token_positions_per_second": 76927.82963734682,
+      "required_pages_per_second_upper_estimate": 0.00532204575146852
     },
     {
       "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": 9646.709532095563,
-      "required_pages_per_second_upper_estimate": 0.0040388758348861705
+      "required_token_positions_per_second": 9615.978704668352,
+      "required_pages_per_second_upper_estimate": 0.004026009479175048
     },
     {
       "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": 19293.419064191126,
-      "required_pages_per_second_upper_estimate": 0.004018310389394858
+      "required_token_positions_per_second": 19231.957409336705,
+      "required_pages_per_second_upper_estimate": 0.004005509547541518
     },
     {
       "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": 28940.12859628669,
-      "required_pages_per_second_upper_estimate": 0.0017349165203296476
+      "required_token_positions_per_second": 28847.936114005057,
+      "required_pages_per_second_upper_estimate": 0.0017293897217866333
     },
     {
       "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": 28940.12859628669,
-      "required_pages_per_second_upper_estimate": 0.002249967704188458
+      "required_token_positions_per_second": 28847.936114005057,
+      "required_pages_per_second_upper_estimate": 0.002242800144202935
     },
     {
       "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": 28940.12859628669,
-      "required_pages_per_second_upper_estimate": 0.3712128961440552
+      "required_token_positions_per_second": 28847.936114005057,
+      "required_pages_per_second_upper_estimate": 0.37003034996992157
     }
   ],
   "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": 0.39438611727348416,
-    "producer_p95_seconds": 0.2660703999972611,
-    "gpu_update_seconds": 0.9553516636255555,
-    "producer_batches_per_second": 2.5355862090514645,
-    "gpu_batches_per_second": 1.046734975270786,
-    "producer_headroom": 2.422376503083332,
+    "producer_batch_seconds": 0.4764918981561266,
+    "producer_p95_seconds": 0.24925029900259688,
+    "gpu_update_seconds": 0.9584047846867453,
+    "producer_batches_per_second": 2.098671570009238,
+    "gpu_batches_per_second": 1.0434004670864097,
+    "producer_headroom": 2.011376874182897,
     "sustainable": true,
-    "buffer_seconds": 60.18715480840999,
+    "buffer_seconds": 60.379501435264956,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315445.8416899396,
-    "upload_bytes_per_second": 5404764.988606308,
+    "generation_bytes_per_second": 286774.6283284259,
+    "upload_bytes_per_second": 5411937.717162252,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.68733442329065,
+    "estimated_full_shards_per_hour": 61.5351594675978,
     "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": 1.1592294109977956,
-      "bytes": 6491968,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.6240387940015353,
       "bytes": 5730323,
@@ -562,6 +556,12 @@
       "bytes": 6027814,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0737022979992616,
+      "bytes": 6591396,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -251,3 +251,4 @@
 {"step": 582200, "loss": 0.6435660719871521, "lr": 0.0003, "grad_norm": 0.18179106690589025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85217344932, "tokens_per_second": 153247.3753550792, "data_wait_s": 7.557002390967682e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 582300, "loss": 0.5519033074378967, "lr": 0.0003, "grad_norm": 0.17627105548430186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85231989464, "tokens_per_second": 153438.6389978105, "data_wait_s": 1.1235999409109354e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 582400, "loss": 0.587166965007782, "lr": 0.0003, "grad_norm": 0.16023193403620564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85246743268, "tokens_per_second": 152593.23711836973, "data_wait_s": 1.982600224437192e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 582500, "loss": 0.6072389483451843, "lr": 0.0003, "grad_norm": 0.1896262142150918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85261442459, "tokens_per_second": 148580.8885633686, "data_wait_s": 8.775001333560795e-06, "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 sha256:44dd195a821c6deaf17f6b62051299127034be324b773dd5212812dfda88287b
-size 539934
+oid sha256:59761138389a54847ca27a01c16610bccb65545ee2aa82542e568f2081011963
+size 542052