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": 74316.58644428951,
-      "required_pages_per_second_upper_estimate": 0.0064415733593246845
+      "required_token_positions_per_second": 74987.97352285849,
+      "required_pages_per_second_upper_estimate": 0.00649976748968005
     },
     {
       "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": 9289.573305536189,
-      "required_pages_per_second_upper_estimate": 0.0037057496830765076
+      "required_token_positions_per_second": 9373.496690357311,
+      "required_pages_per_second_upper_estimate": 0.003739227976048074
     },
     {
       "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": 18579.146611072378,
-      "required_pages_per_second_upper_estimate": 0.004297958586659056
+      "required_token_positions_per_second": 18746.993380714623,
+      "required_pages_per_second_upper_estimate": 0.004336786982813535
     },
     {
       "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": 27868.719916608567,
-      "required_pages_per_second_upper_estimate": 0.0016970588270482174
+      "required_token_positions_per_second": 28120.490071071934,
+      "required_pages_per_second_upper_estimate": 0.0017123903085191233
     },
     {
       "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": 27868.719916608567,
-      "required_pages_per_second_upper_estimate": 0.002145411204734792
+      "required_token_positions_per_second": 28120.490071071934,
+      "required_pages_per_second_upper_estimate": 0.0021647931681697758
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27868.719916608567,
-      "required_pages_per_second_upper_estimate": 0.3460362307586397
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28120.490071071934,
+      "required_pages_per_second_upper_estimate": 0.3491623734565719
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.335111455656147,
-    "producer_p95_seconds": 0.3941534000041429,
-    "gpu_update_seconds": 0.9920800123841703,
-    "producer_batches_per_second": 2.9840818125480184,
-    "gpu_batches_per_second": 1.0079832145764094,
-    "producer_headroom": 2.9604479215480155,
+    "producer_batch_seconds": 0.14352952918733308,
+    "producer_p95_seconds": 0.3995997410020209,
+    "gpu_update_seconds": 0.9831976587222427,
+    "producer_batches_per_second": 6.967207414822712,
+    "gpu_batches_per_second": 1.0170894846307847,
+    "producer_headroom": 6.850142018085941,
     "sustainable": true,
-    "buffer_seconds": 60.51688075543439,
+    "buffer_seconds": 60.95825484077905,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 425627.32299853646,
-    "upload_bytes_per_second": 5718900.184719799,
+    "generation_bytes_per_second": 410385.87319828605,
+    "upload_bytes_per_second": 5698611.475179152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.32971541850156,
+    "estimated_full_shards_per_hour": 88.05925509416042,
     "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": 1.2875092799986305,
-      "bytes": 13122964,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.006686715001706,
       "bytes": 1247241,
@@ -562,6 +556,12 @@
       "bytes": 11695145,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.135505132995604,
+      "bytes": 10769907,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -355,3 +355,4 @@
 {"step": 230500, "loss": 0.5772615559399128, "lr": 0.0003, "grad_norm": 0.17446808019470036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33738817563, "tokens_per_second": 149107.7870497439, "data_wait_s": 1.2322001566644758e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230600, "loss": 0.5562775582075119, "lr": 0.0003, "grad_norm": 0.15923243384969488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33753458501, "tokens_per_second": 148866.19270517267, "data_wait_s": 1.694400270935148e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 230700, "loss": 0.5872198939323425, "lr": 0.0003, "grad_norm": 0.15921065386942876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33768145527, "tokens_per_second": 148661.03372752288, "data_wait_s": 1.469899871153757e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 230800, "loss": 0.6101261675357819, "lr": 0.0003, "grad_norm": 0.1448520216478739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33782797597, "tokens_per_second": 148328.51270472183, "data_wait_s": 7.295006071217358e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:11185f4ed31884924949b5f6d14fbe95d91d784e84acc461b4d1d739a6f8c582
-size 761772
+oid sha256:0135b99d246b872efaf82802cb3a86aac6985c9c06cadd7fb6a7e83b78dddb69
+size 763890