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": 77384.34071182295,
-      "required_pages_per_second_upper_estimate": 0.005353628246852408
+      "required_token_positions_per_second": 77257.56929986458,
+      "required_pages_per_second_upper_estimate": 0.005344857906422927
     },
     {
       "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": 9673.042588977869,
-      "required_pages_per_second_upper_estimate": 0.004049900935906034
+      "required_token_positions_per_second": 9657.196162483073,
+      "required_pages_per_second_upper_estimate": 0.004043266368043677
     },
     {
       "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": 19346.085177955738,
-      "required_pages_per_second_upper_estimate": 0.004029279351993207
+      "required_token_positions_per_second": 19314.392324966146,
+      "required_pages_per_second_upper_estimate": 0.004022678566512214
     },
     {
       "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": 29019.127766933605,
-      "required_pages_per_second_upper_estimate": 0.001739652399985181
+      "required_token_positions_per_second": 28971.58848744922,
+      "required_pages_per_second_upper_estimate": 0.0017368024927683688
     },
     {
       "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": 29019.127766933605,
-      "required_pages_per_second_upper_estimate": 0.002256109542225627
+      "required_token_positions_per_second": 28971.58848744922,
+      "required_pages_per_second_upper_estimate": 0.0022524135723489074
     },
     {
       "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": 29019.127766933605,
-      "required_pages_per_second_upper_estimate": 0.37222621268241307
+      "required_token_positions_per_second": 28971.58848744922,
+      "required_pages_per_second_upper_estimate": 0.37161642984888876
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09306053483612686,
-    "producer_p95_seconds": 0.27246575100070913,
-    "gpu_update_seconds": 0.9527508966518297,
-    "producer_batches_per_second": 10.745693668760131,
-    "gpu_batches_per_second": 1.0495922948109666,
-    "producer_headroom": 10.237969278057104,
+    "producer_batch_seconds": 0.08004848707818724,
+    "producer_p95_seconds": 0.27503556800002116,
+    "gpu_update_seconds": 0.9543142590188795,
+    "producer_batches_per_second": 12.492428483042428,
+    "gpu_batches_per_second": 1.047872847491653,
+    "producer_headroom": 11.92170263114098,
     "sustainable": true,
-    "buffer_seconds": 60.02330648906527,
+    "buffer_seconds": 60.121798318189406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 165,
+    "upload_pages_per_commit": 167,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 664115.8172760308,
+    "generation_bytes_per_second": 672946.671730145,
     "upload_bytes_per_second": 5277695.71805537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.25189728122088,
+    "estimated_full_shards_per_hour": 72.19934517826205,
     "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": [

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

@@ -249,3 +249,4 @@
 {"step": 669900, "loss": 0.5402909517288208, "lr": 0.0001, "grad_norm": 0.1947490227177393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98056439999, "tokens_per_second": 153219.01603945586, "data_wait_s": 1.0189996828557923e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 670000, "loss": 0.5217463970184326, "lr": 0.0001, "grad_norm": 0.18787271377557538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98071134777, "tokens_per_second": 153469.18739889583, "data_wait_s": 9.882998710963875e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 670100, "loss": 0.6055551767349243, "lr": 0.0001, "grad_norm": 0.21865299620761078, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98085959934, "tokens_per_second": 141357.3574487342, "data_wait_s": 8.374998287763447e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 670200, "loss": 0.5137597918510437, "lr": 0.0001, "grad_norm": 0.1884368122650523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98100440735, "tokens_per_second": 150034.85117115034, "data_wait_s": 1.516799966339022e-05, "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:cb96f76159f015621342bd10018893e61680aa54b4cd6a56004ac40eeb694fda
-size 535698
+oid sha256:d7be0b57f1941191c3e6a2119a13190aba2f116096d73d2b2a036dd13e0d2905
+size 537816