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": 76703.28817277803,
-      "required_pages_per_second_upper_estimate": 0.005306511451941673
+      "required_token_positions_per_second": 77454.76677604989,
+      "required_pages_per_second_upper_estimate": 0.005358500485386611
     },
     {
       "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": 9587.911021597254,
-      "required_pages_per_second_upper_estimate": 0.004014258126393052
+      "required_token_positions_per_second": 9681.845847006236,
+      "required_pages_per_second_upper_estimate": 0.004053586676209579
     },
     {
       "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": 19175.82204319451,
-      "required_pages_per_second_upper_estimate": 0.003993818031163256
+      "required_token_positions_per_second": 19363.691694012472,
+      "required_pages_per_second_upper_estimate": 0.004032946324972773
     },
     {
       "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": 28763.733064791766,
-      "required_pages_per_second_upper_estimate": 0.0017243418775569083
+      "required_token_positions_per_second": 29045.537541018708,
+      "required_pages_per_second_upper_estimate": 0.0017412356256161928
     },
     {
       "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": 28763.733064791766,
-      "required_pages_per_second_upper_estimate": 0.002236253727496673
+      "required_token_positions_per_second": 29045.537541018708,
+      "required_pages_per_second_upper_estimate": 0.002258162785996424
     },
     {
       "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": 28763.733064791766,
-      "required_pages_per_second_upper_estimate": 0.36895028366480376
+      "required_token_positions_per_second": 29045.537541018708,
+      "required_pages_per_second_upper_estimate": 0.37256496890777063
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0980609001722712,
-    "producer_p95_seconds": 0.2802222360005544,
-    "gpu_update_seconds": 0.9612104220867813,
-    "producer_batches_per_second": 10.19774444496453,
-    "gpu_batches_per_second": 1.04035492855873,
-    "producer_headroom": 9.802178242277485,
+    "producer_batch_seconds": 0.10953115082830323,
+    "producer_p95_seconds": 0.2707078889980039,
+    "gpu_update_seconds": 0.9518846039931237,
+    "producer_batches_per_second": 9.129822816958812,
+    "gpu_batches_per_second": 1.0505475094407808,
+    "producer_headroom": 8.690537776648226,
     "sustainable": true,
-    "buffer_seconds": 60.556256591467225,
+    "buffer_seconds": 60.920614655559916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 372493.93079099775,
+    "generation_bytes_per_second": 387218.27817679214,
     "upload_bytes_per_second": 6101851.120600281,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.9285263328309,
+    "estimated_full_shards_per_hour": 83.08802851655791,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -243,3 +243,4 @@
 {"step": 509300, "loss": 0.5844889879226685, "lr": 0.0003, "grad_norm": 0.16537971772199328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74543980505, "tokens_per_second": 150588.78921485308, "data_wait_s": 1.4104000001680106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509400, "loss": 0.6208956241607666, "lr": 0.0003, "grad_norm": 0.17751308819122566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74558482958, "tokens_per_second": 150978.5163113281, "data_wait_s": 2.1413001377368346e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 509500, "loss": 0.735479474067688, "lr": 0.0003, "grad_norm": 0.21468405559765869, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74573229358, "tokens_per_second": 153840.5896974438, "data_wait_s": 1.0335003025829792e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 509600, "loss": 0.6320120692253113, "lr": 0.0003, "grad_norm": 0.1685832064861581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74587943237, "tokens_per_second": 152510.95287292535, "data_wait_s": 1.6240999684669077e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a90d863470b8bd10a5b696b81c93730c0d9bd6d99de91acee0498d6f5abd61f
-size 522207
+oid sha256:ebfa234ca72395f9fe8ef24bea545cf4b54a0844eca30caa2ea777122693abef
+size 524325