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": 76228.52706763113,
-      "required_pages_per_second_upper_estimate": 0.005273666377090076
+      "required_token_positions_per_second": 76559.89313560311,
+      "required_pages_per_second_upper_estimate": 0.005296591050547586
     },
     {
       "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": 9528.565883453892,
-      "required_pages_per_second_upper_estimate": 0.003989411556319832
+      "required_token_positions_per_second": 9569.986641950389,
+      "required_pages_per_second_upper_estimate": 0.00400675356293852
     },
     {
       "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": 19057.131766907783,
-      "required_pages_per_second_upper_estimate": 0.003969097976685805
+      "required_token_positions_per_second": 19139.973283900777,
+      "required_pages_per_second_upper_estimate": 0.003986351679997729
     },
     {
       "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": 28585.697650361675,
-      "required_pages_per_second_upper_estimate": 0.0017136689263061618
+      "required_token_positions_per_second": 28709.959925851166,
+      "required_pages_per_second_upper_estimate": 0.0017211182599842483
     },
     {
       "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": 28585.697650361675,
-      "required_pages_per_second_upper_estimate": 0.0022224122571197612
+      "required_token_positions_per_second": 28709.959925851166,
+      "required_pages_per_second_upper_estimate": 0.0022320731024670828
     },
     {
       "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": 28585.697650361675,
-      "required_pages_per_second_upper_estimate": 0.36666663652802906
+      "required_token_positions_per_second": 28709.959925851166,
+      "required_pages_per_second_upper_estimate": 0.368260539575572
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12697045671120577,
-    "producer_p95_seconds": 0.3123786209980608,
-    "gpu_update_seconds": 0.9671969646558614,
-    "producer_batches_per_second": 7.875847861794334,
-    "gpu_batches_per_second": 1.03391556895116,
-    "producer_headroom": 7.617496146018835,
+    "producer_batch_seconds": 0.1092738243827398,
+    "producer_p95_seconds": 0.3248369110006024,
+    "gpu_update_seconds": 0.9630107485836318,
+    "producer_batches_per_second": 9.151322429216211,
+    "gpu_batches_per_second": 1.0384100088921862,
+    "producer_headroom": 8.812821863089683,
     "sustainable": true,
-    "buffer_seconds": 60.933408773319265,
+    "buffer_seconds": 60.6696771607688,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421079.1457939014,
+    "generation_bytes_per_second": 546133.688033724,
     "upload_bytes_per_second": 4583475.731431366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.35378246653349,
+    "estimated_full_shards_per_hour": 117.18757611044683,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -420,3 +420,4 @@
 {"step": 762000, "loss": 0.5618194341659546, "lr": 5e-05, "grad_norm": 0.25727874525994243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111534488578, "tokens_per_second": 152381.51411472692, "data_wait_s": 1.201800478156656e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762100, "loss": 0.5437605977058411, "lr": 5e-05, "grad_norm": 0.2541325931069935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111549176468, "tokens_per_second": 151530.7859179251, "data_wait_s": 1.4659999578725547e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762200, "loss": 0.5626640319824219, "lr": 5e-05, "grad_norm": 0.2490859000687136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111563720181, "tokens_per_second": 150494.9558218023, "data_wait_s": 1.3183998817112297e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 762300, "loss": 0.564041793346405, "lr": 5e-05, "grad_norm": 0.28050232875361786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111578230490, "tokens_per_second": 150239.1350671474, "data_wait_s": 1.6789999790489674e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9d883647259dfd759ce10b04700f95d1a491a87020795a21ecc17c6c74b2555
-size 900225
+oid sha256:80d0be225f0caabb43862dfa743ae3acfc0643243311830aed1959bc3017cd6e
+size 902343