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": 75693.32692111358,
-      "required_pages_per_second_upper_estimate": 0.006560905734538957
+      "required_token_positions_per_second": 75882.00864454966,
+      "required_pages_per_second_upper_estimate": 0.00657726018811694
     },
     {
       "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": 9461.665865139197,
-      "required_pages_per_second_upper_estimate": 0.0037743999781152055
+      "required_token_positions_per_second": 9485.251080568707,
+      "required_pages_per_second_upper_estimate": 0.003783808473180432
     },
     {
       "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": 18923.331730278394,
-      "required_pages_per_second_upper_estimate": 0.004377579756536157
+      "required_token_positions_per_second": 18970.502161137414,
+      "required_pages_per_second_upper_estimate": 0.004388491805544162
     },
     {
       "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": 28384.997595417593,
-      "required_pages_per_second_upper_estimate": 0.0017284974289880457
+      "required_token_positions_per_second": 28455.75324170612,
+      "required_pages_per_second_upper_estimate": 0.0017328060766208307
     },
     {
       "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": 28384.997595417593,
-      "required_pages_per_second_upper_estimate": 0.002185155689597595
+      "required_token_positions_per_second": 28455.75324170612,
+      "required_pages_per_second_upper_estimate": 0.002190602655113051
     },
     {
       "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": 28384.997595417593,
-      "required_pages_per_second_upper_estimate": 0.3524466716592075
+      "required_token_positions_per_second": 28455.75324170612,
+      "required_pages_per_second_upper_estimate": 0.353325219982196
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15083036395307658,
-    "producer_p95_seconds": 0.20434694799769204,
-    "gpu_update_seconds": 0.9740356646873004,
-    "producer_batches_per_second": 6.629964774938159,
-    "gpu_batches_per_second": 1.0266564523805553,
-    "producer_headroom": 6.457822146410278,
+    "producer_batch_seconds": 0.35676528807022123,
+    "producer_p95_seconds": 0.27817409000272164,
+    "gpu_update_seconds": 0.9716137107724762,
+    "producer_batches_per_second": 2.8029632742834907,
+    "gpu_batches_per_second": 1.0292156120408753,
+    "producer_headroom": 2.723397548085552,
     "sustainable": true,
-    "buffer_seconds": 60.39021121061263,
+    "buffer_seconds": 60.24005006789352,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 501655.8319262954,
+    "generation_bytes_per_second": 480011.302384843,
     "upload_bytes_per_second": 5574908.085737823,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.64366358129165,
+    "estimated_full_shards_per_hour": 102.99925140055626,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 288800, "loss": 0.6307291015982628, "lr": 0.0003, "grad_norm": 0.15888960789895745, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42272264646, "tokens_per_second": 152438.4921092194, "data_wait_s": 7.839000318199396e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 288900, "loss": 0.6174089070409536, "lr": 0.0003, "grad_norm": 0.15672216329987154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42286949926, "tokens_per_second": 151189.8196565876, "data_wait_s": 1.2953998520970345e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 289000, "loss": 0.5663933157920837, "lr": 0.0003, "grad_norm": 0.1463761686236363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42301548432, "tokens_per_second": 149546.27295082275, "data_wait_s": 1.3045999367022887e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 289100, "loss": 0.6017055995762348, "lr": 0.0003, "grad_norm": 0.15662319902066163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42316200440, "tokens_per_second": 150184.7106884939, "data_wait_s": 9.663999662734568e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3ca8984dbb1cefa6363919d4367704180868434c611904eefdf058523daefb8b
-size 187332
+oid sha256:ca9f9ffac3555b587eb549cd2d81ed6dba57b38ec849be3d362cf1846254bc75
+size 189450