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": 75936.39264073719,
-      "required_pages_per_second_upper_estimate": 0.006581974055071498
+      "required_token_positions_per_second": 75826.29482896872,
+      "required_pages_per_second_upper_estimate": 0.006572431055787759
     },
     {
       "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": 9492.049080092149,
-      "required_pages_per_second_upper_estimate": 0.0037865202968294836
+      "required_token_positions_per_second": 9478.28685362109,
+      "required_pages_per_second_upper_estimate": 0.0037810303389265556
     },
     {
       "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": 18984.098160184298,
-      "required_pages_per_second_upper_estimate": 0.004391637000642248
+      "required_token_positions_per_second": 18956.57370724218,
+      "required_pages_per_second_upper_estimate": 0.004385269702867916
     },
     {
       "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": 28476.147240276445,
-      "required_pages_per_second_upper_estimate": 0.0017340479641347224
+      "required_token_positions_per_second": 28434.860560863268,
+      "required_pages_per_second_upper_estimate": 0.0017315338219729259
     },
     {
       "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": 28476.147240276445,
-      "required_pages_per_second_upper_estimate": 0.0021921726415771955
+      "required_token_positions_per_second": 28434.860560863268,
+      "required_pages_per_second_upper_estimate": 0.0021889942787071256
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28476.147240276445,
-      "required_pages_per_second_upper_estimate": 0.3535784451901169
+      "required_token_positions_per_second": 28434.860560863268,
+      "required_pages_per_second_upper_estimate": 0.3530658028094325
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08388351715629483,
-    "producer_p95_seconds": 0.3408927730015421,
-    "gpu_update_seconds": 0.9709178621219826,
-    "producer_batches_per_second": 11.921293168200894,
-    "gpu_batches_per_second": 1.0299532421974986,
-    "producer_headroom": 11.574596476599009,
+    "producer_batch_seconds": 0.11058859329691018,
+    "producer_p95_seconds": 0.3290460290008923,
+    "gpu_update_seconds": 0.9723276096543876,
+    "producer_batches_per_second": 9.042523918494764,
+    "gpu_batches_per_second": 1.0284599450543717,
+    "producer_headroom": 8.79229566691264,
     "sustainable": true,
-    "buffer_seconds": 60.19690745156292,
+    "buffer_seconds": 60.28431179857203,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377250.3772586177,
-    "upload_bytes_per_second": 4904947.984668851,
+    "generation_bytes_per_second": 376393.0961548195,
+    "upload_bytes_per_second": 4864077.851231555,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.94914902037524,
+    "estimated_full_shards_per_hour": 80.76519645198286,
     "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": 1.2434416440009954,
-      "bytes": 10283141,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.034428533999744,
       "bytes": 7372684,
@@ -562,6 +556,12 @@
       "bytes": 10372111,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.1448558080010116,
+      "bytes": 6260573,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 268300, "loss": 0.6001504268497229, "lr": 0.0003, "grad_norm": 0.16388218285261316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39271157663, "tokens_per_second": 148489.45665295143, "data_wait_s": 1.3158998626749963e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268400, "loss": 0.6030313111841679, "lr": 0.0003, "grad_norm": 0.15376708163393907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39285573765, "tokens_per_second": 147003.52498788704, "data_wait_s": 1.2368000170681626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 268500, "loss": 0.5589021369814873, "lr": 0.0003, "grad_norm": 0.1745483632741585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39300244150, "tokens_per_second": 149560.46126671965, "data_wait_s": 1.2019001587759703e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 268600, "loss": 0.5550219640135765, "lr": 0.0003, "grad_norm": 0.1480452006594306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39314917054, "tokens_per_second": 149776.29686238288, "data_wait_s": 1.7141002899734303e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dca9f9dc0355fdc71db1efec23cf8231ddb0607e9cd1dde096a59be191ae7e28
-size 669915
+oid sha256:7a6495ecba10271f7b05d5000a96a6b970ada03f5459420e836e5e5db3630382
+size 672033