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": 75827.06668128644,
-      "required_pages_per_second_upper_estimate": 0.006572497957990417
+      "required_token_positions_per_second": 74157.75252182032,
+      "required_pages_per_second_upper_estimate": 0.006427806037485951
     },
     {
       "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": 9478.383335160805,
-      "required_pages_per_second_upper_estimate": 0.0037810688268552755
+      "required_token_positions_per_second": 9269.71906522754,
+      "required_pages_per_second_upper_estimate": 0.0036978295297700418
     },
     {
       "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": 18956.76667032161,
-      "required_pages_per_second_upper_estimate": 0.004385314341480307
+      "required_token_positions_per_second": 18539.43813045508,
+      "required_pages_per_second_upper_estimate": 0.004288772728513634
     },
     {
       "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": 28435.15000548241,
-      "required_pages_per_second_upper_estimate": 0.0017315514476316332
+      "required_token_positions_per_second": 27809.15719568262,
+      "required_pages_per_second_upper_estimate": 0.0016934317698452724
     },
     {
       "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": 28435.15000548241,
-      "required_pages_per_second_upper_estimate": 0.0021890165609551415
+      "required_token_positions_per_second": 27809.15719568262,
+      "required_pages_per_second_upper_estimate": 0.002140825901597748
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28435.15000548241,
-      "required_pages_per_second_upper_estimate": 0.35306939674289345
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27809.15719568262,
+      "required_pages_per_second_upper_estimate": 0.345296661108343
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.3755156574997045,
-    "producer_p95_seconds": 0.36393893199783633,
-    "gpu_update_seconds": 0.9723177122212949,
-    "producer_batches_per_second": 2.6630048042691454,
-    "gpu_batches_per_second": 1.0284704139714413,
-    "producer_headroom": 2.5892867389212926,
+    "producer_batch_seconds": 0.2583969681016356,
+    "producer_p95_seconds": 0.3476230289998057,
+    "gpu_update_seconds": 0.9942048874567245,
+    "producer_batches_per_second": 3.8700144484925567,
+    "gpu_batches_per_second": 1.005828891626252,
+    "producer_headroom": 3.84758727921944,
     "sustainable": true,
-    "buffer_seconds": 60.28369815772029,
+    "buffer_seconds": 60.646498134860195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387372.5956148947,
+    "generation_bytes_per_second": 380360.10351027007,
     "upload_bytes_per_second": 5468217.38674446,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.12114144644862,
+    "estimated_full_shards_per_hour": 81.61642388325764,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 181400, "loss": 0.5794285256415606, "lr": 0.0003, "grad_norm": 0.15885148057593385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26549235565, "tokens_per_second": 150434.89883062337, "data_wait_s": 1.758099824655801e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 181500, "loss": 0.6017955429852009, "lr": 0.0003, "grad_norm": 0.1433503516335405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26563937611, "tokens_per_second": 149179.17245515654, "data_wait_s": 9.309002052759752e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 181600, "loss": 0.5649946480989456, "lr": 0.0003, "grad_norm": 0.13432296823074194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26578620979, "tokens_per_second": 149636.97172450728, "data_wait_s": 7.533999450970441e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 181700, "loss": 0.6719467882066965, "lr": 0.0003, "grad_norm": 0.15690616034954094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26593333780, "tokens_per_second": 149614.2150639295, "data_wait_s": 1.3469998521031812e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb1f7ebc63e0a6639b86b159f6a6c0100cad4db10eaaabad8dcdd09f71f8bf88
-size 567468
+oid sha256:7c955ecfd833c84b0f045702565b01632f5610340646b706671afc9fac83b1f6
+size 569586