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": 77387.31565487587,
-      "required_pages_per_second_upper_estimate": 0.005353834060315628
+      "required_token_positions_per_second": 75747.19480319506,
+      "required_pages_per_second_upper_estimate": 0.005240366694191664
     },
     {
       "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": 9673.414456859484,
-      "required_pages_per_second_upper_estimate": 0.004050056629222582
+      "required_token_positions_per_second": 9468.399350399382,
+      "required_pages_per_second_upper_estimate": 0.003964221085349991
     },
     {
       "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": 19346.828913718968,
-      "required_pages_per_second_upper_estimate": 0.004029434252539057
+      "required_token_positions_per_second": 18936.798700798765,
+      "required_pages_per_second_upper_estimate": 0.003944035772411651
     },
     {
       "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": 29020.243370578453,
-      "required_pages_per_second_upper_estimate": 0.0017397192787202615
+      "required_token_positions_per_second": 28405.198051198146,
+      "required_pages_per_second_upper_estimate": 0.001702848250943238
     },
     {
       "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": 29020.243370578453,
-      "required_pages_per_second_upper_estimate": 0.002256196275501992
+      "required_token_positions_per_second": 28405.198051198146,
+      "required_pages_per_second_upper_estimate": 0.002208379207218628
     },
     {
       "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": 29020.243370578453,
-      "required_pages_per_second_upper_estimate": 0.37224052244812733
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28405.198051198146,
+      "required_pages_per_second_upper_estimate": 0.36435138147532925
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 704566,
-    "producer_batch_seconds": 0.09065078686718575,
-    "producer_p95_seconds": 0.1295661530000416,
-    "gpu_update_seconds": 0.9527142707572993,
-    "producer_batches_per_second": 11.03134384773868,
-    "gpu_batches_per_second": 1.0496326450585378,
-    "producer_headroom": 10.509718709371379,
+    "producer_batch_seconds": 0.04240800200000905,
+    "producer_p95_seconds": 0.12717883700020138,
+    "gpu_update_seconds": 0.9733429758231801,
+    "producer_batches_per_second": 23.580455405557345,
+    "gpu_batches_per_second": 1.0273870822915996,
+    "producer_headroom": 22.95187063571098,
     "sustainable": true,
-    "buffer_seconds": 60.02099905770986,
+    "buffer_seconds": 60.347264501037166,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 109,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 55,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 818335.0990835065,
-    "upload_bytes_per_second": 2507271.1629311224,
+    "generation_bytes_per_second": 333996.60020538163,
+    "upload_bytes_per_second": 4346409.302599536,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.7978311985917,
+    "estimated_full_shards_per_hour": 71.66789536114776,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 1226596,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.2994146069999033,
+      "bytes": 17142614,
+      "pages": 34,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -4,3 +4,4 @@
 {"step": 557500, "loss": 0.6515524983406067, "lr": 0.0003, "grad_norm": 0.2547269670709619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81600042352, "tokens_per_second": 151738.7602942602, "data_wait_s": 1.4462999956776912e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557600, "loss": 0.6659453511238098, "lr": 0.0003, "grad_norm": 0.20856943782813392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81614775357, "tokens_per_second": 151948.99221986908, "data_wait_s": 1.635600006011373e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 557700, "loss": 0.6180841326713562, "lr": 0.0003, "grad_norm": 0.15532756566934822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81629633185, "tokens_per_second": 118041.87280050917, "data_wait_s": 9.15299995085661e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 557800, "loss": 0.6057742834091187, "lr": 0.0003, "grad_norm": 0.15931434321151466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81644154641, "tokens_per_second": 150808.6348368384, "data_wait_s": 1.4784999848416192e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48feedb735ee40cdab5296da374ee1cd39f36c2266891789d41386a1f8502d5d
-size 12873
+oid sha256:af954c927de95e3f7724889e370f1d36055937f27a29d125984c593973a46c71
+size 14991