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": 77485.59919485156,
-      "required_pages_per_second_upper_estimate": 0.005360633543660378
+      "required_token_positions_per_second": 77162.27315944084,
+      "required_pages_per_second_upper_estimate": 0.005338265098310903
     },
     {
       "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": 9685.699899356445,
-      "required_pages_per_second_upper_estimate": 0.004055200287446847
+      "required_token_positions_per_second": 9645.284144930105,
+      "required_pages_per_second_upper_estimate": 0.004038279055045463
     },
     {
       "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": 19371.39979871289,
-      "required_pages_per_second_upper_estimate": 0.004034551719905479
+      "required_token_positions_per_second": 19290.56828986021,
+      "required_pages_per_second_upper_estimate": 0.004017716648281703
     },
     {
       "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": 29057.099698069334,
-      "required_pages_per_second_upper_estimate": 0.0017419287592769863
+      "required_token_positions_per_second": 28935.852434790315,
+      "required_pages_per_second_upper_estimate": 0.0017346601709772604
     },
     {
       "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": 29057.099698069334,
-      "required_pages_per_second_upper_estimate": 0.002259061692850556
+      "required_token_positions_per_second": 28935.852434790315,
+      "required_pages_per_second_upper_estimate": 0.0022496352514409616
     },
     {
       "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": 29057.099698069334,
-      "required_pages_per_second_upper_estimate": 0.3727132758439391
+      "required_token_positions_per_second": 28935.852434790315,
+      "required_pages_per_second_upper_estimate": 0.37115804613576425
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05960329035937306,
-    "producer_p95_seconds": 0.13803009599996585,
-    "gpu_update_seconds": 0.9515058380667304,
-    "producer_batches_per_second": 16.77759724288011,
-    "gpu_batches_per_second": 1.0509657008850308,
-    "producer_headroom": 15.963981725332703,
+    "producer_batch_seconds": 0.06965682292186592,
+    "producer_p95_seconds": 0.1390849460001391,
+    "gpu_update_seconds": 0.9554928461951272,
+    "producer_batches_per_second": 14.356095469954182,
+    "gpu_batches_per_second": 1.0465803108648117,
+    "producer_headroom": 13.717146520835493,
     "sustainable": true,
-    "buffer_seconds": 60.896373636270745,
+    "buffer_seconds": 60.19604931029301,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 67,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422029.3303230604,
-    "upload_bytes_per_second": 5817129.37474625,
+    "generation_bytes_per_second": 349190.123807557,
+    "upload_bytes_per_second": 6108620.20311586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.55766994732723,
+    "estimated_full_shards_per_hour": 74.92807183904678,
     "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": [
@@ -232,6 +232,12 @@
       "bytes": 1220702,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.144875766000041,
+      "bytes": 9911722,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -12,3 +12,4 @@
 {"step": 441400, "loss": 0.5791196227073669, "lr": 0.0003, "grad_norm": 0.14784625682046532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64607772031, "tokens_per_second": 152950.68091202807, "data_wait_s": 1.7840000055002747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441500, "loss": 0.6097682118415833, "lr": 0.0003, "grad_norm": 0.14591231422891635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64622581565, "tokens_per_second": 153395.51456268365, "data_wait_s": 1.4266000107454602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 441600, "loss": 0.5996808409690857, "lr": 0.0003, "grad_norm": 0.15014320042516954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64637254119, "tokens_per_second": 138165.69149354866, "data_wait_s": 1.3468000133798341e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 441700, "loss": 0.5900413393974304, "lr": 0.0003, "grad_norm": 0.15800119962399603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64651752436, "tokens_per_second": 150390.5225978341, "data_wait_s": 1.516900010756217e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:19c8b2d54894ca4de84f3a68312fb260d5302c0f0cbe0443e936e5b8210cbe5b
-size 29817
+oid sha256:ccc5c9e53fb87cf66796b23ed37173460d92816995e32a73ba8c7339b78170c4
+size 31935