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": 77133.54554362336,
-      "required_pages_per_second_upper_estimate": 0.005336277655191397
+      "required_token_positions_per_second": 77513.93857340144,
+      "required_pages_per_second_upper_estimate": 0.005362594127624877
     },
     {
       "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": 9641.69319295292,
-      "required_pages_per_second_upper_estimate": 0.0040367755984402195
+      "required_token_positions_per_second": 9689.24232167518,
+      "required_pages_per_second_upper_estimate": 0.00405668342569751
     },
     {
       "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": 19283.38638590584,
-      "required_pages_per_second_upper_estimate": 0.004016220847087551
+      "required_token_positions_per_second": 19378.48464335036,
+      "required_pages_per_second_upper_estimate": 0.004036027306203546
     },
     {
       "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": 28925.07957885876,
-      "required_pages_per_second_upper_estimate": 0.0017340143547133623
+      "required_token_positions_per_second": 29067.726965025537,
+      "required_pages_per_second_upper_estimate": 0.0017425658477041135
     },
     {
       "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": 28925.07957885876,
-      "required_pages_per_second_upper_estimate": 0.0022487977092770684
+      "required_token_positions_per_second": 29067.726965025537,
+      "required_pages_per_second_upper_estimate": 0.0022598879161119935
     },
     {
       "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": 28925.07957885876,
-      "required_pages_per_second_upper_estimate": 0.3710198635068657
+      "required_token_positions_per_second": 29067.726965025537,
+      "required_pages_per_second_upper_estimate": 0.372849591013783
     }
   ],
   "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": 0.04669425489838375,
-    "producer_p95_seconds": 0.15238068899998325,
-    "gpu_update_seconds": 0.9558487099274163,
-    "producer_batches_per_second": 21.415910847623646,
-    "gpu_batches_per_second": 1.0461906676381207,
-    "producer_headroom": 20.470370755621623,
+    "producer_batch_seconds": 0.10410337956246707,
+    "producer_p95_seconds": 0.15692672499972105,
+    "gpu_update_seconds": 0.9511579640632458,
+    "producer_batches_per_second": 9.605836085272829,
+    "gpu_batches_per_second": 1.0513500783067684,
+    "producer_headroom": 9.136667493993363,
     "sustainable": true,
-    "buffer_seconds": 60.21846872542723,
+    "buffer_seconds": 60.87410970004773,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 492778.4313396868,
-    "upload_bytes_per_second": 2812955.1245735097,
+    "generation_bytes_per_second": 507224.9812672838,
+    "upload_bytes_per_second": 2906244.8475536816,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.7387800707145,
+    "estimated_full_shards_per_hour": 108.8386733867062,
     "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": [
@@ -274,6 +274,12 @@
       "bytes": 6174644,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.431994874999873,
+      "bytes": 7632897,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 528500, "loss": 0.644488513469696, "lr": 0.0003, "grad_norm": 0.17564630383103438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77354750227, "tokens_per_second": 153199.0555196711, "data_wait_s": 1.1514000107126776e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 528600, "loss": 0.5591337084770203, "lr": 0.0003, "grad_norm": 0.14620017999408466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77369414809, "tokens_per_second": 152438.932562098, "data_wait_s": 9.891999980027322e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 528700, "loss": 0.564262866973877, "lr": 0.0003, "grad_norm": 0.1696123026690492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77384149819, "tokens_per_second": 153880.50341834704, "data_wait_s": 7.919999916339293e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 528800, "loss": 0.6091707348823547, "lr": 0.0003, "grad_norm": 0.19470882718978458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77398908528, "tokens_per_second": 152481.44281903357, "data_wait_s": 1.6463000065414235e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:54041a9cac36f88c159a000168d19c47ea4cfba46653b32866bb77ab43cb0720
-size 78531
+oid sha256:1030b37520de693dcfd17c441eca738f6ceef8bbfae4c37a9b0bd0e49c0b6859
+size 80649