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": 77281.5042804526,
-      "required_pages_per_second_upper_estimate": 0.005346513783916812
+      "required_token_positions_per_second": 77497.54145697228,
+      "required_pages_per_second_upper_estimate": 0.005361459737063755
     },
     {
       "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": 9660.188035056575,
-      "required_pages_per_second_upper_estimate": 0.004044519002612799
+      "required_token_positions_per_second": 9687.192682121535,
+      "required_pages_per_second_upper_estimate": 0.004055825284417741
     },
     {
       "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": 19320.37607011315,
-      "required_pages_per_second_upper_estimate": 0.004023924822824363
+      "required_token_positions_per_second": 19374.38536424307,
+      "required_pages_per_second_upper_estimate": 0.004035173534470758
     },
     {
       "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": 28980.564105169728,
-      "required_pages_per_second_upper_estimate": 0.0017373405673457387
+      "required_token_positions_per_second": 29061.578046364604,
+      "required_pages_per_second_upper_estimate": 0.0017421972294192458
     },
     {
       "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": 28980.564105169728,
-      "required_pages_per_second_upper_estimate": 0.0022531113871471073
+      "required_token_positions_per_second": 29061.578046364604,
+      "required_pages_per_second_upper_estimate": 0.002259409864732341
     },
     {
       "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": 28980.564105169728,
-      "required_pages_per_second_upper_estimate": 0.37173155943574004
+      "required_token_positions_per_second": 29061.578046364604,
+      "required_pages_per_second_upper_estimate": 0.372770719287395
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0925905562891387,
-    "producer_p95_seconds": 0.2639257139999245,
-    "gpu_update_seconds": 0.9540186967950698,
-    "producer_batches_per_second": 10.800237519658412,
-    "gpu_batches_per_second": 1.0481974864427708,
-    "producer_headroom": 10.303628523581736,
+    "producer_batch_seconds": 0.1396459998594537,
+    "producer_p95_seconds": 0.27141610000035143,
+    "gpu_update_seconds": 0.9513592123555922,
+    "producer_batches_per_second": 7.160964159420585,
+    "gpu_batches_per_second": 1.0511276781815901,
+    "producer_headroom": 6.812649222412993,
     "sustainable": true,
-    "buffer_seconds": 60.1031778980894,
+    "buffer_seconds": 60.8869895907579,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448098.7984056709,
-    "upload_bytes_per_second": 5569697.290323856,
+    "generation_bytes_per_second": 461108.29557427915,
+    "upload_bytes_per_second": 5486491.221233017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.1515709316978,
+    "estimated_full_shards_per_hour": 98.94310617848664,
     "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": 1.493546093999612,
-      "bytes": 10942153,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 1.185240850001719,
       "bytes": 5667842,
@@ -562,6 +556,12 @@
       "bytes": 10630556,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1581052289984655,
+      "bytes": 2607511,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -221,3 +221,4 @@
 {"step": 507100, "loss": 0.6167461276054382, "lr": 0.0003, "grad_norm": 0.17118224313750252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74222571305, "tokens_per_second": 150993.34753739194, "data_wait_s": 1.2050000805174932e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507200, "loss": 0.5564034581184387, "lr": 0.0003, "grad_norm": 0.16956993233946016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74237249256, "tokens_per_second": 153147.2302347653, "data_wait_s": 7.195001671789214e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 507300, "loss": 0.5764183402061462, "lr": 0.0003, "grad_norm": 0.15764874691966407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74251910837, "tokens_per_second": 153151.64194877763, "data_wait_s": 1.147299917647615e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 507400, "loss": 0.5982879996299744, "lr": 0.0003, "grad_norm": 0.1681867520757786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74266776156, "tokens_per_second": 154267.11232209133, "data_wait_s": 1.3694003428099677e-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 sha256:7a663c614557b2c0414507ed0d27286c1c768cc7f37f7283060a2fc0b96cb00b
-size 475611
+oid sha256:339342a06aa55e9360eb1e3cf65a0b01afa0a6ba8abd606f0e3318c6260da3e9
+size 477729