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": 76060.43532710502,
-      "required_pages_per_second_upper_estimate": 0.005262037400453907
+      "required_token_positions_per_second": 76702.22020501335,
+      "required_pages_per_second_upper_estimate": 0.005306437567453686
     },
     {
       "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": 9507.554415888128,
-      "required_pages_per_second_upper_estimate": 0.003980614493619384
+      "required_token_positions_per_second": 9587.777525626669,
+      "required_pages_per_second_upper_estimate": 0.004014202234417881
     },
     {
       "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": 19015.108831776255,
-      "required_pages_per_second_upper_estimate": 0.003960345707517232
+      "required_token_positions_per_second": 19175.555051253337,
+      "required_pages_per_second_upper_estimate": 0.003993762423782961
     },
     {
       "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": 28522.663247664383,
-      "required_pages_per_second_upper_estimate": 0.0017098901101124191
+      "required_token_positions_per_second": 28763.332576880006,
+      "required_pages_per_second_upper_estimate": 0.0017243178689181078
     },
     {
       "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": 28522.663247664383,
-      "required_pages_per_second_upper_estimate": 0.002217511609569083
+      "required_token_positions_per_second": 28763.332576880006,
+      "required_pages_per_second_upper_estimate": 0.0022362225913230937
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28522.663247664383,
-      "required_pages_per_second_upper_estimate": 0.3658580988912967
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28763.332576880006,
+      "required_pages_per_second_upper_estimate": 0.3689451466358821
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.517374471320295,
-    "producer_p95_seconds": 0.2349287680008274,
-    "gpu_update_seconds": 0.9693344467846632,
-    "producer_batches_per_second": 1.9328359929474028,
-    "gpu_batches_per_second": 1.0316356788073056,
-    "producer_headroom": 1.8735645079491559,
+    "producer_batch_seconds": 0.11650432875796923,
+    "producer_p95_seconds": 0.22508133299925248,
+    "gpu_update_seconds": 0.9612238055552537,
+    "producer_batches_per_second": 8.583372057165706,
+    "gpu_batches_per_second": 1.0403404433188659,
+    "producer_headroom": 8.250541553285446,
     "sustainable": true,
-    "buffer_seconds": 60.098735700649115,
+    "buffer_seconds": 60.55709974998098,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409520.1887909491,
-    "upload_bytes_per_second": 1847065.6762338295,
+    "generation_bytes_per_second": 393998.6829313322,
+    "upload_bytes_per_second": 1908164.8147142662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.87349937244754,
+    "estimated_full_shards_per_hour": 84.54294553713774,
     "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": 1.157564569999522,
-      "bytes": 2379174,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1466768219997903,
       "bytes": 6100233,
@@ -562,6 +556,12 @@
       "bytes": 972942,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.5192261520023749,
+      "bytes": 17166243,
+      "pages": 34,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 661900, "loss": 0.533353328704834, "lr": 0.0001, "grad_norm": 0.18042469287622923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96885118580, "tokens_per_second": 151442.23393889933, "data_wait_s": 1.3823999324813485e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 662000, "loss": 0.5269763469696045, "lr": 0.0001, "grad_norm": 0.1846541889624236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96899719797, "tokens_per_second": 152010.6874511778, "data_wait_s": 1.4915000065229833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 662100, "loss": 0.5512187480926514, "lr": 0.0001, "grad_norm": 0.20899591529591302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96914305363, "tokens_per_second": 151980.4583637133, "data_wait_s": 1.154500205302611e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 662200, "loss": 0.5222975611686707, "lr": 0.0001, "grad_norm": 0.17397302274548612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96928982551, "tokens_per_second": 153071.21664682732, "data_wait_s": 8.088998583843932e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f219557a5754e764c238c40623de30bf2fdaf13a8f1b4564dd82b8514a366d52
-size 364692
+oid sha256:536f7745324860cf1deff9a206669eba1ed98242f4e055d10a3640bad074ba0e
+size 366810