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": 77442.5236969844,
-      "required_pages_per_second_upper_estimate": 0.005357653480769003
+      "required_token_positions_per_second": 77094.58464003397,
+      "required_pages_per_second_upper_estimate": 0.005333582249479331
     },
     {
       "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": 9680.31546212305,
-      "required_pages_per_second_upper_estimate": 0.0040529459360170595
+      "required_token_positions_per_second": 9636.823080004246,
+      "required_pages_per_second_upper_estimate": 0.004034736583848133
     },
     {
       "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": 19360.6309242461,
-      "required_pages_per_second_upper_estimate": 0.004032308847348364
+      "required_token_positions_per_second": 19273.64616000849,
+      "required_pages_per_second_upper_estimate": 0.004014192214900164
     },
     {
       "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": 29040.946386369153,
-      "required_pages_per_second_upper_estimate": 0.0017409603929052393
+      "required_token_positions_per_second": 28910.46924001274,
+      "required_pages_per_second_upper_estimate": 0.0017331384872082392
     },
     {
       "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": 29040.946386369153,
-      "required_pages_per_second_upper_estimate": 0.002257805843916776
+      "required_token_positions_per_second": 28910.46924001274,
+      "required_pages_per_second_upper_estimate": 0.0022476618196958795
     },
     {
       "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": 29040.946386369153,
-      "required_pages_per_second_upper_estimate": 0.372506078505524
+      "required_token_positions_per_second": 28910.46924001274,
+      "required_pages_per_second_upper_estimate": 0.3708324577675086
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0772843979218436,
-    "producer_p95_seconds": 0.19702451099874452,
-    "gpu_update_seconds": 0.9520350897716283,
-    "producer_batches_per_second": 12.939222234884756,
-    "gpu_batches_per_second": 1.0503814520532824,
-    "producer_headroom": 12.318593601963558,
+    "producer_batch_seconds": 0.11080308114834736,
+    "producer_p95_seconds": 0.21257938199960336,
+    "gpu_update_seconds": 0.9563317623961132,
+    "producer_batches_per_second": 9.025019788584778,
+    "gpu_batches_per_second": 1.0456622265629607,
+    "producer_headroom": 8.630913080077079,
     "sustainable": true,
-    "buffer_seconds": 60.93024574538421,
+    "buffer_seconds": 60.24890103095513,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 526695.3265453463,
-    "upload_bytes_per_second": 5727778.815449332,
+    "generation_bytes_per_second": 505405.7150191804,
+    "upload_bytes_per_second": 5791207.642921905,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.01655623693745,
+    "estimated_full_shards_per_hour": 108.44830120021399,
     "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": 0.9477791999997862,
-      "bytes": 2414897,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.0987587250001525,
       "bytes": 11899150,
@@ -562,6 +556,12 @@
       "bytes": 1113077,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.0412252780006384,
+      "bytes": 7665353,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -117,3 +117,4 @@
 {"step": 536700, "loss": 0.659512996673584, "lr": 0.0003, "grad_norm": 0.15999194652839047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78555528991, "tokens_per_second": 152619.0411828723, "data_wait_s": 1.0633999409037642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536800, "loss": 0.5641829967498779, "lr": 0.0003, "grad_norm": 0.1685009959452132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78569947244, "tokens_per_second": 151059.9830976834, "data_wait_s": 1.779000012902543e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 536900, "loss": 0.5877827405929565, "lr": 0.0003, "grad_norm": 0.18093387534735209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78584399217, "tokens_per_second": 150909.51258987506, "data_wait_s": 1.4298000678536482e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 537000, "loss": 0.6134182214736938, "lr": 0.0003, "grad_norm": 0.16458220191705777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78599136816, "tokens_per_second": 154497.98174207684, "data_wait_s": 1.591399995959364e-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 sha256:170e6c7f3aa4985a146f70458eb09d816eef8f0d6a44091157d327af6a34093b
-size 253773
+oid sha256:be3815ade7dd2b17af2a58770b2225663c6b32d17c119393dc05d743ed658b3e
+size 255891