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": 75501.3095122896,
-      "required_pages_per_second_upper_estimate": 0.005223355779234346
+      "required_token_positions_per_second": 77108.5678047361,
+      "required_pages_per_second_upper_estimate": 0.005334549637258837
     },
     {
       "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": 9437.6636890362,
-      "required_pages_per_second_upper_estimate": 0.003951352705770822
+      "required_token_positions_per_second": 9638.570975592012,
+      "required_pages_per_second_upper_estimate": 0.00403546839123052
     },
     {
       "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": 18875.3273780724,
-      "required_pages_per_second_upper_estimate": 0.003931232916995545
+      "required_token_positions_per_second": 19277.141951184025,
+      "required_pages_per_second_upper_estimate": 0.004014920296011815
     },
     {
       "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": 28312.991067108604,
-      "required_pages_per_second_upper_estimate": 0.0016973205830389825
+      "required_token_positions_per_second": 28915.712926776036,
+      "required_pages_per_second_upper_estimate": 0.0017334528382230527
     },
     {
       "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": 28312.991067108604,
-      "required_pages_per_second_upper_estimate": 0.002201210519781327
+      "required_token_positions_per_second": 28915.712926776036,
+      "required_pages_per_second_upper_estimate": 0.0022480694932771853
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28312.991067108604,
-      "required_pages_per_second_upper_estimate": 0.3631686492875746
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28915.712926776036,
+      "required_pages_per_second_upper_estimate": 0.370899718151076
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3967515599530316,
-    "producer_p95_seconds": 0.26873104799960856,
-    "gpu_update_seconds": 0.9765128641642836,
-    "producer_batches_per_second": 2.520468980937044,
-    "gpu_batches_per_second": 1.0240520495916017,
-    "producer_headroom": 2.461270383612066,
+    "producer_batch_seconds": 0.09170077097664375,
+    "producer_p95_seconds": 0.25694418599960045,
+    "gpu_update_seconds": 0.9561583375106021,
+    "producer_batches_per_second": 10.905033723813519,
+    "gpu_batches_per_second": 1.045851885372397,
+    "producer_headroom": 10.426938915858583,
     "sustainable": true,
-    "buffer_seconds": 60.54379757818558,
+    "buffer_seconds": 60.23797526316793,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290529.505912798,
-    "upload_bytes_per_second": 4783958.9458460165,
+    "generation_bytes_per_second": 306689.1680943889,
+    "upload_bytes_per_second": 4779566.796762844,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.340868788127466,
+    "estimated_full_shards_per_hour": 65.80835611461401,
     "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": 1.3341312570009904,
-      "bytes": 10704029,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.136130429997138,
       "bytes": 965107,
@@ -562,6 +556,12 @@
       "bytes": 7165592,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.7035883720018319,
+      "bytes": 12069614,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -269,3 +269,4 @@
 {"step": 911900, "loss": 0.5535483956336975, "lr": 3e-05, "grad_norm": 0.3318998368641891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133475655567, "tokens_per_second": 152193.08449663408, "data_wait_s": 1.050200080499053e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912000, "loss": 0.554841160774231, "lr": 3e-05, "grad_norm": 0.32315140536455805, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133490399358, "tokens_per_second": 153303.78350690423, "data_wait_s": 1.52729990077205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 912100, "loss": 0.5340381860733032, "lr": 3e-05, "grad_norm": 0.33642669476817466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133505129415, "tokens_per_second": 153514.19093113282, "data_wait_s": 1.6370999219361693e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 912200, "loss": 0.56758713722229, "lr": 3e-05, "grad_norm": 0.33254980483321406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 133519863314, "tokens_per_second": 153573.38938487993, "data_wait_s": 1.5778998204041272e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e95af9d8b56beb2d9cf9680b203aa3c1dbd113b92707686fbbbc06bc2a6944e3
-size 578058
+oid sha256:440a9edb5ae87c7c7ed9c1989f6ab127f8c8bfdca99c196d0aca6a446d7c197b
+size 580176