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": 76722.23201940082,
-      "required_pages_per_second_upper_estimate": 0.005307822031206818
+      "required_token_positions_per_second": 74857.84871464508,
+      "required_pages_per_second_upper_estimate": 0.005178839668218547
     },
     {
       "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": 9590.279002425103,
-      "required_pages_per_second_upper_estimate": 0.004015249550516609
+      "required_token_positions_per_second": 9357.231089330635,
+      "required_pages_per_second_upper_estimate": 0.00391767725589778
     },
     {
       "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": 19180.558004850205,
-      "required_pages_per_second_upper_estimate": 0.0039948044070804295
+      "required_token_positions_per_second": 18714.46217866127,
+      "required_pages_per_second_upper_estimate": 0.003897728938258797
     },
     {
       "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": 28770.837007275306,
-      "required_pages_per_second_upper_estimate": 0.0017247677480617043
+      "required_token_positions_per_second": 28071.693267991905,
+      "required_pages_per_second_upper_estimate": 0.0016828551484223424
     },
     {
       "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": 28770.837007275306,
-      "required_pages_per_second_upper_estimate": 0.0022368060277778274
+      "required_token_positions_per_second": 28071.693267991905,
+      "required_pages_per_second_upper_estimate": 0.002182450677256854
     },
     {
       "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": 28770.837007275306,
-      "required_pages_per_second_upper_estimate": 0.36904140541136343
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28071.693267991905,
+      "required_pages_per_second_upper_estimate": 0.3600735402058966
     }
   ],
   "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": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10299202926574935,
-    "producer_p95_seconds": 0.25948805200096103,
-    "gpu_update_seconds": 0.9609730851072781,
-    "producer_batches_per_second": 9.709489240373248,
-    "gpu_batches_per_second": 1.0406118709228627,
-    "producer_headroom": 9.330557830137401,
+    "producer_batch_seconds": 0.5195123940390545,
+    "producer_p95_seconds": 0.2837376419993234,
+    "gpu_update_seconds": 0.9849067434605019,
+    "producer_batches_per_second": 1.9248818920859563,
+    "gpu_batches_per_second": 1.0153245539638276,
+    "producer_headroom": 1.8958291558804683,
     "sustainable": true,
-    "buffer_seconds": 60.54130436175852,
+    "buffer_seconds": 60.07931135109062,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398614.7902798205,
+    "generation_bytes_per_second": 358350.7912028812,
     "upload_bytes_per_second": 3562801.1227366845,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.53345471664392,
+    "estimated_full_shards_per_hour": 76.89373781266048,
     "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": [

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 864500, "loss": 0.5429225564002991, "lr": 3e-05, "grad_norm": 0.3118571488107704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126538749991, "tokens_per_second": 153312.32491908857, "data_wait_s": 1.8873997760238126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 864600, "loss": 0.5103281140327454, "lr": 3e-05, "grad_norm": 0.31034518389707766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126553375745, "tokens_per_second": 151861.51612500154, "data_wait_s": 9.679999493528157e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 864700, "loss": 0.5366424918174744, "lr": 3e-05, "grad_norm": 0.30245920492425526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126568088983, "tokens_per_second": 152589.92227156385, "data_wait_s": 1.617299858480692e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 864800, "loss": 0.5435219407081604, "lr": 3e-05, "grad_norm": 0.32043555136016144, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126582788786, "tokens_per_second": 152000.72178230458, "data_wait_s": 1.5872999938437715e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62874d7ff6d78703c97c1cef936f8579bbf93db5a094633d6391984adaa91e75
-size 473493
+oid sha256:880794d8756895500f66cae2ec2b9aa36267c072913ba9e6baaa7f35a9b4543d
+size 475611