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": 77469.76852772056,
-      "required_pages_per_second_upper_estimate": 0.005359538341376048
+      "required_token_positions_per_second": 76987.9890876563,
+      "required_pages_per_second_upper_estimate": 0.0053262077218300865
     },
     {
       "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": 9683.72106596507,
-      "required_pages_per_second_upper_estimate": 0.004054371791228618
+      "required_token_positions_per_second": 9623.498635957038,
+      "required_pages_per_second_upper_estimate": 0.004029157917371598
     },
     {
       "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": 19367.44213193014,
-      "required_pages_per_second_upper_estimate": 0.004033727442285324
+      "required_token_positions_per_second": 19246.997271914075,
+      "required_pages_per_second_upper_estimate": 0.0040086419542885365
     },
     {
       "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": 29051.163197895214,
-      "required_pages_per_second_upper_estimate": 0.0017415728751560588
+      "required_token_positions_per_second": 28870.495907871114,
+      "required_pages_per_second_upper_estimate": 0.001730742146982093
     },
     {
       "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": 29051.163197895214,
-      "required_pages_per_second_upper_estimate": 0.002258600156073923
+      "required_token_positions_per_second": 28870.495907871114,
+      "required_pages_per_second_upper_estimate": 0.0022445540689459745
     },
     {
       "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": 29051.163197895214,
-      "required_pages_per_second_upper_estimate": 0.3726371287938227
+      "required_token_positions_per_second": 28870.495907871114,
+      "required_pages_per_second_upper_estimate": 0.37031972278281594
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.20061996485937073,
-    "producer_p95_seconds": 0.16659473500021704,
-    "gpu_update_seconds": 0.9517002748448684,
-    "producer_batches_per_second": 4.984548774599644,
-    "gpu_batches_per_second": 1.0507509837201683,
-    "producer_headroom": 4.743796438764133,
+    "producer_batch_seconds": 0.07956431094529393,
+    "producer_p95_seconds": 0.16687191200071538,
+    "gpu_update_seconds": 0.9576558742956051,
+    "producer_batches_per_second": 12.568449196871327,
+    "gpu_batches_per_second": 1.0442164318529772,
+    "producer_headroom": 12.036249204169707,
     "sustainable": true,
-    "buffer_seconds": 60.90881759007158,
+    "buffer_seconds": 60.33232008062312,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467363.2133265804,
-    "upload_bytes_per_second": 1859928.1602896368,
+    "generation_bytes_per_second": 450003.18101693864,
+    "upload_bytes_per_second": 1928027.8113718212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.28526592109736,
+    "estimated_full_shards_per_hour": 96.56020710831756,
     "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": [
@@ -280,6 +280,12 @@
       "bytes": 4940477,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.1281410550000146,
+      "bytes": 6022620,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -52,3 +52,4 @@
 {"step": 650200, "loss": 0.5408124923706055, "lr": 0.0001, "grad_norm": 0.18690006406968004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95172613887, "tokens_per_second": 153961.18191900602, "data_wait_s": 6.558000677614473e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 650300, "loss": 0.6387020349502563, "lr": 0.0001, "grad_norm": 0.19528463559114845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95187083881, "tokens_per_second": 151118.3570929203, "data_wait_s": 1.4895000276737846e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 650400, "loss": 0.4960200786590576, "lr": 0.0001, "grad_norm": 0.1797841009292938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95201569260, "tokens_per_second": 150431.6185882629, "data_wait_s": 1.4076000297791325e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 650500, "loss": 0.533645749092102, "lr": 0.0001, "grad_norm": 0.17229197103661847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95216212674, "tokens_per_second": 152768.75228491038, "data_wait_s": 8.107999747153372e-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:d250c55a152d1bded6dd5b8396be65e7dfcb479e6b1d9aad0c006ef451f38084
-size 115320
+oid sha256:c4ad5168a16255488e3cfc5123ba77f13a7f61cd69c50400d2cf465e25e7ad09
+size 117438