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": 76830.0436360616,
-      "required_pages_per_second_upper_estimate": 0.005315280689004826
+      "required_token_positions_per_second": 76460.9210057776,
+      "required_pages_per_second_upper_estimate": 0.005289743928959277
     },
     {
       "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": 9603.7554545077,
-      "required_pages_per_second_upper_estimate": 0.004020891859583272
+      "required_token_positions_per_second": 9557.6151257222,
+      "required_pages_per_second_upper_estimate": 0.004001573867440414
     },
     {
       "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": 19207.5109090154,
-      "required_pages_per_second_upper_estimate": 0.004000417986222158
+      "required_token_positions_per_second": 19115.2302514444,
+      "required_pages_per_second_upper_estimate": 0.003981198358854712
     },
     {
       "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": 28811.266363523104,
-      "required_pages_per_second_upper_estimate": 0.0017271914262367073
+      "required_token_positions_per_second": 28672.8453771666,
+      "required_pages_per_second_upper_estimate": 0.0017188932994612431
     },
     {
       "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": 28811.266363523104,
-      "required_pages_per_second_upper_estimate": 0.0022399492323961494
+      "required_token_positions_per_second": 28672.8453771666,
+      "required_pages_per_second_upper_estimate": 0.0022291876095565048
     },
     {
       "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": 28811.266363523104,
-      "required_pages_per_second_upper_estimate": 0.36955998978364957
+      "required_token_positions_per_second": 28672.8453771666,
+      "required_pages_per_second_upper_estimate": 0.3677844739955439
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.14660934570252948,
-    "producer_p95_seconds": 0.3278310200039414,
-    "gpu_update_seconds": 0.9596246014026002,
-    "producer_batches_per_second": 6.820847574267203,
-    "gpu_batches_per_second": 1.0420741595602974,
-    "producer_headroom": 6.545453134684057,
+    "producer_batch_seconds": 0.12579964089070472,
+    "producer_p95_seconds": 0.3255883210003958,
+    "gpu_update_seconds": 0.964257283723131,
+    "producer_batches_per_second": 7.949148287862001,
+    "gpu_batches_per_second": 1.0370676134681207,
+    "producer_headroom": 7.66502413596619,
     "sustainable": true,
-    "buffer_seconds": 60.45634988836381,
+    "buffer_seconds": 60.74820887455726,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 332827.753042944,
-    "upload_bytes_per_second": 3841901.405224155,
+    "generation_bytes_per_second": 332634.1685259543,
+    "upload_bytes_per_second": 3763751.7201299695,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.41708796945801,
+    "estimated_full_shards_per_hour": 71.37554923852893,
     "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": 1.5923971939992043,
-      "bytes": 18139508,
-      "pages": 43,
-      "shards": 5
-    },
     {
       "seconds": 0.9787014109970187,
       "bytes": 5956252,
@@ -562,6 +556,12 @@
       "bytes": 5294252,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1802646279975306,
+      "bytes": 7575689,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -393,3 +393,4 @@
 {"step": 524300, "loss": 0.5188298225402832, "lr": 0.0003, "grad_norm": 0.18324355156873937, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76739694092, "tokens_per_second": 152502.48113458435, "data_wait_s": 8.791997970547527e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 524400, "loss": 0.629214346408844, "lr": 0.0003, "grad_norm": 0.1776136555103492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76754444777, "tokens_per_second": 154079.28275669058, "data_wait_s": 1.5911005903035402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 524500, "loss": 0.6034004092216492, "lr": 0.0003, "grad_norm": 0.1641275472761367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76769183056, "tokens_per_second": 153989.75978307574, "data_wait_s": 9.735995263326913e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 524600, "loss": 0.5592507123947144, "lr": 0.0003, "grad_norm": 0.15388680595723428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76783892664, "tokens_per_second": 154076.70086355964, "data_wait_s": 1.5998994058463722e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:63ee80e9bd9bbbdf4048315c55eed01185f806397159cc7001851cc385021a4c
-size 842256
+oid sha256:d1144498eda558f467cf8feb6fe8c440849c957520fa1e53c4a6383f8d805fe4
+size 844374