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": 75912.4010115279,
-      "required_pages_per_second_upper_estimate": 0.005251796043015346
+      "required_token_positions_per_second": 76086.17770826326,
+      "required_pages_per_second_upper_estimate": 0.005263818318112987
     },
     {
       "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": 9489.050126440987,
-      "required_pages_per_second_upper_estimate": 0.00397286713404137
+      "required_token_positions_per_second": 9510.772213532908,
+      "required_pages_per_second_upper_estimate": 0.0039819617182979975
     },
     {
       "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": 18978.100252881974,
-      "required_pages_per_second_upper_estimate": 0.00395263779651541
+      "required_token_positions_per_second": 19021.544427065815,
+      "required_pages_per_second_upper_estimate": 0.003961686072297986
     },
     {
       "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": 28467.150379322964,
-      "required_pages_per_second_upper_estimate": 0.0017065621984185932
+      "required_token_positions_per_second": 28532.316640598725,
+      "required_pages_per_second_upper_estimate": 0.0017104688162789554
     },
     {
       "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": 28467.150379322964,
-      "required_pages_per_second_upper_estimate": 0.00221319572823084
+      "required_token_positions_per_second": 28532.316640598725,
+      "required_pages_per_second_upper_estimate": 0.0022182621184089397
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28467.150379322964,
-      "required_pages_per_second_upper_estimate": 0.3651460394212871
+      "required_token_positions_per_second": 28532.316640598725,
+      "required_pages_per_second_upper_estimate": 0.3659819222508527
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.32535608852336395,
-    "producer_p95_seconds": 0.26996372400026303,
-    "gpu_update_seconds": 0.9712247145074995,
-    "producer_batches_per_second": 3.073555514324391,
-    "gpu_batches_per_second": 1.0296278349002808,
-    "producer_headroom": 2.9851130769226573,
+    "producer_batch_seconds": 0.08963982473433418,
+    "producer_p95_seconds": 0.2474102119995223,
+    "gpu_update_seconds": 0.9690064900183946,
+    "producer_batches_per_second": 11.15575585922555,
+    "gpu_batches_per_second": 1.031984832197581,
+    "producer_headroom": 10.809999828650291,
     "sustainable": true,
-    "buffer_seconds": 60.21593229946497,
+    "buffer_seconds": 60.078402381140464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363092.35891469644,
-    "upload_bytes_per_second": 4874168.134340046,
+    "generation_bytes_per_second": 377772.1447261471,
+    "upload_bytes_per_second": 4875676.556592378,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.91116786556883,
+    "estimated_full_shards_per_hour": 81.06110817278204,
     "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": 0.9403941730006409,
-      "bytes": 2445089,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.042657115000111,
       "bytes": 6054602,
@@ -562,6 +556,12 @@
       "bytes": 11016252,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.6974678700007644,
+      "bytes": 6266676,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 733700, "loss": 0.5465325117111206, "lr": 5e-05, "grad_norm": 0.22246921809149486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107393768791, "tokens_per_second": 151113.96318122078, "data_wait_s": 1.57409995154012e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 733800, "loss": 0.5684446692466736, "lr": 5e-05, "grad_norm": 0.2369546057857919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107408502476, "tokens_per_second": 150686.9003358025, "data_wait_s": 1.927700031956192e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 733900, "loss": 0.48412445187568665, "lr": 5e-05, "grad_norm": 0.2331970975359158, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107423007783, "tokens_per_second": 149102.8416541148, "data_wait_s": 1.6247999155893922e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 734000, "loss": 0.5088242292404175, "lr": 5e-05, "grad_norm": 0.22126292743866083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107437593647, "tokens_per_second": 148952.50170638604, "data_wait_s": 2.1376999939093366e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57ad91fcfd2fc72d1dde08cba0dbc051d4d043d71dbd5811f14ce63346c3e376
-size 296133
+oid sha256:eacc7289ff5768a2746007c2fc80f5cebda3f49cf86c9883eb83d8dff478396c
+size 298251