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": 75440.28773213664,
-      "required_pages_per_second_upper_estimate": 0.006538972938960397
+      "required_token_positions_per_second": 75345.47845031362,
+      "required_pages_per_second_upper_estimate": 0.006530755110703906
     },
     {
       "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": 9430.03596651708,
-      "required_pages_per_second_upper_estimate": 0.0037617823386457156
+      "required_token_positions_per_second": 9418.184806289202,
+      "required_pages_per_second_upper_estimate": 0.0037570547336401795
     },
     {
       "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": 18860.07193303416,
-      "required_pages_per_second_upper_estimate": 0.0043629457157781095
+      "required_token_positions_per_second": 18836.369612578404,
+      "required_pages_per_second_upper_estimate": 0.004357462601087264
     },
     {
       "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": 28290.10789955124,
-      "required_pages_per_second_upper_estimate": 0.0017227191443575432
+      "required_token_positions_per_second": 28254.55441886761,
+      "required_pages_per_second_upper_estimate": 0.0017205541239186072
     },
     {
       "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": 28290.10789955124,
-      "required_pages_per_second_upper_estimate": 0.0021778508181383134
+      "required_token_positions_per_second": 28254.55441886761,
+      "required_pages_per_second_upper_estimate": 0.002175113812776954
     },
     {
       "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": 28290.10789955124,
-      "required_pages_per_second_upper_estimate": 0.3512684592119304
+      "required_token_positions_per_second": 28254.55441886761,
+      "required_pages_per_second_upper_estimate": 0.3508270039716852
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13371293042945354,
-    "producer_p95_seconds": 0.35925924600087455,
-    "gpu_update_seconds": 0.9773027412326898,
-    "producer_batches_per_second": 7.47870828040521,
-    "gpu_batches_per_second": 1.0232243887279817,
-    "producer_headroom": 7.308962103319628,
+    "producer_batch_seconds": 0.41054172471856987,
+    "producer_p95_seconds": 0.34582075299840653,
+    "gpu_update_seconds": 0.9785325080737225,
+    "producer_batches_per_second": 2.435806008964154,
+    "gpu_batches_per_second": 1.0219384555435334,
+    "producer_headroom": 2.3835153631327377,
     "sustainable": true,
-    "buffer_seconds": 60.592769956426764,
+    "buffer_seconds": 60.669015500570794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454209.3440592655,
-    "upload_bytes_per_second": 4137910.589778355,
+    "generation_bytes_per_second": 422613.58071665955,
+    "upload_bytes_per_second": 4150722.990893746,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.46275178273653,
+    "estimated_full_shards_per_hour": 90.68303648114052,
     "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": 1.0568676230031997,
-      "bytes": 6659990,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1445812689999002,
       "bytes": 6216290,
@@ -562,6 +556,12 @@
       "bytes": 6202267,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.093655648001004,
+      "bytes": 8128170,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -229,3 +229,4 @@
 {"step": 107900, "loss": 0.602509293705225, "lr": 0.0003, "grad_norm": 0.1348617034635483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15789525209, "tokens_per_second": 148541.37369743414, "data_wait_s": 1.6844001947902143e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108000, "loss": 0.6069594826549292, "lr": 0.0003, "grad_norm": 0.13489505758698042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15804196367, "tokens_per_second": 149057.3969863826, "data_wait_s": 1.3618002412840724e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108100, "loss": 0.6172971874475479, "lr": 0.0003, "grad_norm": 0.13241357751297037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15818680956, "tokens_per_second": 146717.90665073448, "data_wait_s": 1.6654001228744164e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 108200, "loss": 0.6171077489852905, "lr": 0.0003, "grad_norm": 0.141955425435858, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15833177836, "tokens_per_second": 147043.92650415594, "data_wait_s": 1.4761000784346834e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd13e858857da0dd54184a2a4003b9348a27f3346d6f0c891b2efbf078ba8435
-size 492555
+oid sha256:343c8cc02922fb1aa1eddad48c5b7ff7a6c000f261715a9bea7de6987c8f6943
+size 494673