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": 76724.49319933544,
-      "required_pages_per_second_upper_estimate": 0.005307978464881355
+      "required_token_positions_per_second": 76410.14907096534,
+      "required_pages_per_second_upper_estimate": 0.0052862314087018455
     },
     {
       "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": 9590.56164991693,
-      "required_pages_per_second_upper_estimate": 0.004015367889119086
+      "required_token_positions_per_second": 9551.268633870668,
+      "required_pages_per_second_upper_estimate": 0.003998916723831998
     },
     {
       "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": 19181.12329983386,
-      "required_pages_per_second_upper_estimate": 0.003994922143117694
+      "required_token_positions_per_second": 19102.537267741336,
+      "required_pages_per_second_upper_estimate": 0.003978554745085854
     },
     {
       "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": 28771.684949750786,
-      "required_pages_per_second_upper_estimate": 0.0017248185809183763
+      "required_token_positions_per_second": 28653.805901612006,
+      "required_pages_per_second_upper_estimate": 0.001717751911973341
     },
     {
       "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": 28771.684949750786,
-      "required_pages_per_second_upper_estimate": 0.0022368719515755914
+      "required_token_positions_per_second": 28653.805901612006,
+      "required_pages_per_second_upper_estimate": 0.0022277073740779377
     },
     {
       "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": 28771.684949750786,
-      "required_pages_per_second_upper_estimate": 0.36905228190698924
+      "required_token_positions_per_second": 28653.805901612006,
+      "required_pages_per_second_upper_estimate": 0.3675402560461257
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.461310970695024,
-    "producer_p95_seconds": 0.3019174809960532,
-    "gpu_update_seconds": 0.9609447638637333,
-    "producer_batches_per_second": 2.1677351364381643,
-    "gpu_batches_per_second": 1.0406425401385557,
-    "producer_headroom": 2.0830737288036896,
+    "producer_batch_seconds": 0.11051638735182223,
+    "producer_p95_seconds": 0.3054088930002763,
+    "gpu_update_seconds": 0.9648979997608129,
+    "producer_batches_per_second": 9.048431856685294,
+    "gpu_batches_per_second": 1.0363789750293695,
+    "producer_headroom": 8.73081379948766,
     "sustainable": true,
-    "buffer_seconds": 60.539520123415194,
+    "buffer_seconds": 60.78857398493121,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315841.17332939414,
-    "upload_bytes_per_second": 5018151.790426663,
+    "generation_bytes_per_second": 315483.0388251365,
+    "upload_bytes_per_second": 5082445.944608722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.77216339026802,
+    "estimated_full_shards_per_hour": 67.6953160625989,
     "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": 1.2330543320022116,
-      "bytes": 6359348,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.232538223001029,
       "bytes": 5896962,
@@ -562,6 +556,12 @@
       "bytes": 11832847,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2940340749992174,
+      "bytes": 12187860,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 922500, "loss": 0.5190991759300232, "lr": 3e-05, "grad_norm": 0.3377101964376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135027664212, "tokens_per_second": 151365.4228573738, "data_wait_s": 1.276100374525413e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922600, "loss": 0.5873844027519226, "lr": 3e-05, "grad_norm": 0.3420149008279359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135042279656, "tokens_per_second": 151919.74790768157, "data_wait_s": 1.654799416428432e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922700, "loss": 0.5533462166786194, "lr": 3e-05, "grad_norm": 0.33757326006272087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135056896493, "tokens_per_second": 152503.15898803255, "data_wait_s": 8.676994184497744e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 922800, "loss": 0.5286610722541809, "lr": 3e-05, "grad_norm": 0.32214467890983767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135071503388, "tokens_per_second": 152790.25286520587, "data_wait_s": 1.4129996998235583e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f2f65e1036dfbf448f16d9ecd66ba9f54125a0434f50e9ac2c04db1aec7adb48
-size 804132
+oid sha256:9738e4d746b8b6df27e13592f649fd30ff5b5c245d018e8aaf2a3cafb2dfd05a
+size 806250