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": 77588.94080350095,
-      "required_pages_per_second_upper_estimate": 0.005367782955932311
+      "required_token_positions_per_second": 76105.10890189983,
+      "required_pages_per_second_upper_estimate": 0.0052651280220151865
     },
     {
       "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": 9698.61760043762,
-      "required_pages_per_second_upper_estimate": 0.0040606086591372615
+      "required_token_positions_per_second": 9513.138612737479,
+      "required_pages_per_second_upper_estimate": 0.003982952480228916
     },
     {
       "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": 19397.23520087524,
-      "required_pages_per_second_upper_estimate": 0.0040399325528505245
+      "required_token_positions_per_second": 19026.277225474958,
+      "required_pages_per_second_upper_estimate": 0.003962671789394323
     },
     {
       "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": 29095.852801312856,
-      "required_pages_per_second_upper_estimate": 0.001744251948643873
+      "required_token_positions_per_second": 28539.415838212437,
+      "required_pages_per_second_upper_estimate": 0.0017108944023360502
     },
     {
       "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": 29095.852801312856,
-      "required_pages_per_second_upper_estimate": 0.002262074576170853
+      "required_token_positions_per_second": 28539.415838212437,
+      "required_pages_per_second_upper_estimate": 0.0022188140497973355
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29095.852801312856,
-      "required_pages_per_second_upper_estimate": 0.3732103590425066
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28539.415838212437,
+      "required_pages_per_second_upper_estimate": 0.36607298313531683
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1354510863360474,
-    "producer_p95_seconds": 0.32891975599341094,
-    "gpu_update_seconds": 0.9502385164236352,
-    "producer_batches_per_second": 7.38273886943254,
-    "gpu_batches_per_second": 1.0523673611585957,
-    "producer_headroom": 7.015362830432682,
+    "producer_batch_seconds": 0.26059401744555544,
+    "producer_p95_seconds": 0.3880254010000499,
+    "gpu_update_seconds": 0.9687654490454255,
+    "producer_batches_per_second": 3.837386636126153,
+    "gpu_batches_per_second": 1.0322416029446049,
+    "producer_headroom": 3.7175275877076674,
     "sustainable": true,
-    "buffer_seconds": 60.815265051112654,
+    "buffer_seconds": 60.06345784081638,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535572.1608211058,
-    "upload_bytes_per_second": 5179585.356765111,
+    "generation_bytes_per_second": 516578.6940422316,
+    "upload_bytes_per_second": 5193989.985630985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.92131823038942,
+    "estimated_full_shards_per_hour": 110.84576240492068,
     "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": 1.1044605290007894,
-      "bytes": 9770514,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1610946500004502,
       "bytes": 7427357,
@@ -562,6 +556,12 @@
       "bytes": 6311746,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5925084610062186,
+      "bytes": 13513629,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -371,3 +371,4 @@
 {"step": 682100, "loss": 0.5400379300117493, "lr": 0.0001, "grad_norm": 0.18625527652859342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99841673341, "tokens_per_second": 151728.03817951164, "data_wait_s": 1.9319995772093534e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682200, "loss": 0.5104553699493408, "lr": 0.0001, "grad_norm": 0.18129928727462233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99856152253, "tokens_per_second": 151424.87195643186, "data_wait_s": 1.1705997167155147e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 682300, "loss": 0.5576037764549255, "lr": 0.0001, "grad_norm": 0.22132294949022457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99870791671, "tokens_per_second": 151616.09495380416, "data_wait_s": 1.7245998606085777e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 682400, "loss": 0.5416658520698547, "lr": 0.0001, "grad_norm": 0.19516915073326702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99885489902, "tokens_per_second": 152872.42853994438, "data_wait_s": 9.359006071463227e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb303dd3651d1ae64436ed031b1542ac1e0d9dd0cbacd513975a0de70be525c2
-size 795660
+oid sha256:2b37fa5aacc28de8301feb84b639cf4c10f750c387f6bae718e98c01c98440e0
+size 797778