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": 76424.52969193821,
-      "required_pages_per_second_upper_estimate": 0.005287226293428386
+      "required_token_positions_per_second": 75274.40391502832,
+      "required_pages_per_second_upper_estimate": 0.005207657923522288
     },
     {
       "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": 9553.066211492276,
-      "required_pages_per_second_upper_estimate": 0.003999669332044475
+      "required_token_positions_per_second": 9409.30048937854,
+      "required_pages_per_second_upper_estimate": 0.00393947762636512
     },
     {
       "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": 19106.132422984552,
-      "required_pages_per_second_upper_estimate": 0.003979303521112396
+      "required_token_positions_per_second": 18818.60097875708,
+      "required_pages_per_second_upper_estimate": 0.003919418303993913
     },
     {
       "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": 28659.198634476827,
-      "required_pages_per_second_upper_estimate": 0.0017180751980743606
+      "required_token_positions_per_second": 28227.90146813562,
+      "required_pages_per_second_upper_estimate": 0.001692219591504847
     },
     {
       "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": 28659.198634476827,
-      "required_pages_per_second_upper_estimate": 0.0022281266353380525
+      "required_token_positions_per_second": 28227.90146813562,
+      "required_pages_per_second_upper_estimate": 0.002194595177730769
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28659.198634476827,
-      "required_pages_per_second_upper_estimate": 0.3676094282330502
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28227.90146813562,
+      "required_pages_per_second_upper_estimate": 0.3620772112740424
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.49587670881254553,
-    "producer_p95_seconds": 0.28044745400256943,
-    "gpu_update_seconds": 0.9647164372118778,
-    "producer_batches_per_second": 2.016630307954283,
-    "gpu_batches_per_second": 1.0365740246844917,
-    "producer_headroom": 1.945476405863148,
+    "producer_batch_seconds": 0.12319046276556378,
+    "producer_p95_seconds": 0.27727980200143065,
+    "gpu_update_seconds": 0.9794564442280547,
+    "producer_batches_per_second": 8.117511514694435,
+    "gpu_batches_per_second": 1.0209744454620813,
+    "producer_headroom": 7.9507489641629014,
     "sustainable": true,
-    "buffer_seconds": 60.777135544348305,
+    "buffer_seconds": 60.72629954213939,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 255551.76626154265,
-    "upload_bytes_per_second": 3932112.793229657,
+    "generation_bytes_per_second": 266992.64987460274,
+    "upload_bytes_per_second": 3907647.365923125,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.83546009907445,
+    "estimated_full_shards_per_hour": 57.290407392297375,
     "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": 1.4686280049991183,
-      "bytes": 8470485,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.9930366190001223,
       "bytes": 5866239,
@@ -562,6 +556,12 @@
       "bytes": 5696136,
       "pages": 26,
       "shards": 3
+    },
+    {
+      "seconds": 1.6552812369991443,
+      "bytes": 6589515,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -212,3 +212,4 @@
 {"step": 821200, "loss": 0.5458189249038696, "lr": 3e-05, "grad_norm": 0.2782306516437237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120200872674, "tokens_per_second": 151976.70244240065, "data_wait_s": 1.0406998626422137e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821300, "loss": 0.48540323972702026, "lr": 3e-05, "grad_norm": 0.2919788006193103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120215390016, "tokens_per_second": 149984.20810734577, "data_wait_s": 1.160899773822166e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 821400, "loss": 0.5347105264663696, "lr": 3e-05, "grad_norm": 0.2932911313652176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120229858489, "tokens_per_second": 149093.66529550718, "data_wait_s": 1.1603999155340716e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 821500, "loss": 0.43291744589805603, "lr": 3e-05, "grad_norm": 0.2694788604155506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120244409128, "tokens_per_second": 150553.77588452472, "data_wait_s": 1.385100040351972e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0cee412ec6b4665b9c1f676eaf226a98e9150b569b64af5f6f146931cd7db53d
-size 456549
+oid sha256:8aedd6ea2c4cf11512028bbb552a11cd8b66db56edeb316f0406ab74d328fcfe
+size 458667