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": 75825.26510830206,
-      "required_pages_per_second_upper_estimate": 0.0065723418022101765
+      "required_token_positions_per_second": 75935.05639032192,
+      "required_pages_per_second_upper_estimate": 0.0065818582322721975
     },
     {
       "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": 9478.158138537758,
-      "required_pages_per_second_upper_estimate": 0.0037809789925553525
+      "required_token_positions_per_second": 9491.88204879024,
+      "required_pages_per_second_upper_estimate": 0.003786453665545811
     },
     {
       "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": 18956.316277075515,
-      "required_pages_per_second_upper_estimate": 0.004385210150929474
+      "required_token_positions_per_second": 18983.76409758048,
+      "required_pages_per_second_upper_estimate": 0.004391559721138155
     },
     {
       "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": 28434.474415613273,
-      "required_pages_per_second_upper_estimate": 0.0017315103077531488
+      "required_token_positions_per_second": 28475.64614637072,
+      "required_pages_per_second_upper_estimate": 0.0017340174501449003
     },
     {
       "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": 28434.474415613273,
-      "required_pages_per_second_upper_estimate": 0.0021889645521768627
+      "required_token_positions_per_second": 28475.64614637072,
+      "required_pages_per_second_upper_estimate": 0.0021921340659812162
     },
     {
       "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": 28434.474415613273,
-      "required_pages_per_second_upper_estimate": 0.35306100817777264
+      "required_token_positions_per_second": 28475.64614637072,
+      "required_pages_per_second_upper_estimate": 0.35357222328086124
     }
   ],
   "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": 0.07938274952334723,
-    "producer_p95_seconds": 0.28919509800107335,
-    "gpu_update_seconds": 0.9723408140372933,
-    "producer_batches_per_second": 12.597195310120751,
-    "gpu_batches_per_second": 1.0284459785739755,
-    "producer_headroom": 12.248767142429585,
+    "producer_batch_seconds": 0.11323472352367503,
+    "producer_p95_seconds": 0.3285827990002872,
+    "gpu_update_seconds": 0.9709349476350265,
+    "producer_batches_per_second": 8.831213331756144,
+    "gpu_batches_per_second": 1.0299351181413021,
+    "producer_headroom": 8.5745336538224,
     "sustainable": true,
-    "buffer_seconds": 60.28513047031219,
+    "buffer_seconds": 60.197966753371645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 388231.38346823276,
-    "upload_bytes_per_second": 4810537.81939834,
+    "generation_bytes_per_second": 390108.4003287658,
+    "upload_bytes_per_second": 4834260.257281728,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.3054173282169,
+    "estimated_full_shards_per_hour": 83.70818145177107,
     "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": 0.8563921040004061,
-      "bytes": 1024816,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.4308035579997522,
       "bytes": 7307496,
@@ -562,6 +556,12 @@
       "bytes": 7578437,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.5002685959989321,
+      "bytes": 6193769,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -318,3 +318,4 @@
 {"step": 268900, "loss": 0.5969801694154739, "lr": 0.0003, "grad_norm": 0.14266062221497894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39358934556, "tokens_per_second": 150935.2554989693, "data_wait_s": 1.1551001080079004e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269000, "loss": 0.5755130127072334, "lr": 0.0003, "grad_norm": 0.15571733407282995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39373457926, "tokens_per_second": 147619.7950301421, "data_wait_s": 1.1454998457338661e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 269100, "loss": 0.6036871932446957, "lr": 0.0003, "grad_norm": 0.1737280390749632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39388186973, "tokens_per_second": 150143.84551069635, "data_wait_s": 1.0422001651022583e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 269200, "loss": 0.5914702340960503, "lr": 0.0003, "grad_norm": 0.14244819826348482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39402884253, "tokens_per_second": 150400.86541815, "data_wait_s": 1.3136999768903479e-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 sha256:26d12922ea5aca86f725e0dd459745ebcdefee7662db86590221ec9d95f98631
-size 682623
+oid sha256:995df93f7c4caa0a2ae48f7ed4d305af3d2357a5464ef16e2a5c6bd8d83d67a0
+size 684741