Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74645.55466537435,
-      "required_pages_per_second_upper_estimate": 0.006470087490965985
+      "required_token_positions_per_second": 75138.66965150743,
+      "required_pages_per_second_upper_estimate": 0.006512829448175494
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9330.694333171794,
-      "required_pages_per_second_upper_estimate": 0.003722153475814502
+      "required_token_positions_per_second": 9392.333706438429,
+      "required_pages_per_second_upper_estimate": 0.00374674234340132
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18661.388666343588,
-      "required_pages_per_second_upper_estimate": 0.004316983838735312
+      "required_token_positions_per_second": 18784.667412876857,
+      "required_pages_per_second_upper_estimate": 0.004345502207113947
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27992.08299951538,
-      "required_pages_per_second_upper_estimate": 0.0017045709915611672
+      "required_token_positions_per_second": 28177.001119315286,
+      "required_pages_per_second_upper_estimate": 0.0017158315348665873
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27992.08299951538,
-      "required_pages_per_second_upper_estimate": 0.002154908036347828
+      "required_token_positions_per_second": 28177.001119315286,
+      "required_pages_per_second_upper_estimate": 0.002169143545096143
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27992.08299951538,
-      "required_pages_per_second_upper_estimate": 0.34756798737866296
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28177.001119315286,
+      "required_pages_per_second_upper_estimate": 0.3498640515454423
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11103015125006266,
-    "producer_p95_seconds": 0.21583235300022352,
-    "gpu_update_seconds": 0.9877078458390776,
-    "producer_batches_per_second": 9.006562530458911,
-    "gpu_batches_per_second": 1.0124451316375644,
-    "producer_headroom": 8.895852475374523,
+    "producer_batch_seconds": 0.08438231020305409,
+    "producer_p95_seconds": 0.20955442000013136,
+    "gpu_update_seconds": 0.9812257835006915,
+    "producer_batches_per_second": 11.850825102958684,
+    "gpu_batches_per_second": 1.0191334316881975,
+    "producer_headroom": 11.628335146780296,
     "sustainable": true,
-    "buffer_seconds": 60.25017859618373,
+    "buffer_seconds": 60.83599857704287,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355235.4656397556,
-    "upload_bytes_per_second": 5722848.071088951,
+    "generation_bytes_per_second": 355429.276873342,
+    "upload_bytes_per_second": 5803178.402819171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.22526146788121,
+    "estimated_full_shards_per_hour": 76.26684884691424,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 5844217,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.3119218800002272,
+      "bytes": 12680620,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 364300, "loss": 0.5770778506994247, "lr": 0.0003, "grad_norm": 0.15973130305345476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53323134573, "tokens_per_second": 149729.65350880558, "data_wait_s": 7.982000170159154e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 364400, "loss": 0.5820131599903107, "lr": 0.0003, "grad_norm": 0.15329101438276424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53337827905, "tokens_per_second": 148692.86855481096, "data_wait_s": 1.6865000361576676e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 364500, "loss": 0.6504499409347773, "lr": 0.0003, "grad_norm": 0.17116215467605977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53352481057, "tokens_per_second": 149452.50364889574, "data_wait_s": 9.172001227852888e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 364600, "loss": 0.585701584815979, "lr": 0.0003, "grad_norm": 0.14848347318623806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 53367190260, "tokens_per_second": 150535.86390228407, "data_wait_s": 1.3957998817204498e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:65ebbc88480142a606d0517e2f9cd53ed91edad1b4a0e08ec55f3a0a169491c2
-size 202158
+oid sha256:624fe1445ceea15bcefd1ef4424d68027ed8dcf1eb44e58fd08e5a6dbf346ea3
+size 204276