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": 75286.30489288199,
-      "required_pages_per_second_upper_estimate": 0.006525626096719743
+      "required_token_positions_per_second": 75280.57789998564,
+      "required_pages_per_second_upper_estimate": 0.006525129695490418
     },
     {
       "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": 9410.788111610249,
-      "required_pages_per_second_upper_estimate": 0.0037541040815423045
+      "required_token_positions_per_second": 9410.072237498205,
+      "required_pages_per_second_upper_estimate": 0.0037538185086557385
     },
     {
       "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": 18821.576223220498,
-      "required_pages_per_second_upper_estimate": 0.004354040410814072
+      "required_token_positions_per_second": 18820.14447499641,
+      "required_pages_per_second_upper_estimate": 0.0043537092010603375
     },
     {
       "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": 28232.36433483075,
-      "required_pages_per_second_upper_estimate": 0.0017192028642231357
+      "required_token_positions_per_second": 28230.216712494617,
+      "required_pages_per_second_upper_estimate": 0.001719072085290569
     },
     {
       "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": 28232.36433483075,
-      "required_pages_per_second_upper_estimate": 0.0021734055586817122
+      "required_token_positions_per_second": 28230.216712494617,
+      "required_pages_per_second_upper_estimate": 0.0021732402287692777
     },
     {
       "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": 28232.36433483075,
-      "required_pages_per_second_upper_estimate": 0.35055147739338127
+      "required_token_positions_per_second": 28230.216712494617,
+      "required_pages_per_second_upper_estimate": 0.3505248111115961
     }
   ],
   "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": 0.08469448812510905,
-    "producer_p95_seconds": 0.25536394200025825,
-    "gpu_update_seconds": 0.979301615412004,
-    "producer_batches_per_second": 11.807143795742876,
-    "gpu_batches_per_second": 1.0211358628049316,
-    "producer_headroom": 11.56275499257282,
+    "producer_batch_seconds": 0.09845994503106681,
+    "producer_p95_seconds": 0.21742722999988473,
+    "gpu_update_seconds": 0.9793761160807196,
+    "producer_batches_per_second": 10.156414364079449,
+    "gpu_batches_per_second": 1.0210581854924268,
+    "producer_headroom": 9.946949653198564,
     "sustainable": true,
-    "buffer_seconds": 60.716700155544245,
+    "buffer_seconds": 60.72131919700462,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305394.6290508686,
-    "upload_bytes_per_second": 5243727.95480107,
+    "generation_bytes_per_second": 305990.6018114682,
+    "upload_bytes_per_second": 5255244.493300824,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.53057817120117,
+    "estimated_full_shards_per_hour": 65.65846005209002,
     "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": 0.894821537000098,
-      "bytes": 1079304,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.1567702649990679,
       "bytes": 10579644,
@@ -562,6 +556,12 @@
       "bytes": 11269048,
       "pages": 45,
       "shards": 3
+    },
+    {
+      "seconds": 1.0205925500013109,
+      "bytes": 2657105,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -170,3 +170,4 @@
 {"step": 212000, "loss": 0.6270451061427593, "lr": 0.0003, "grad_norm": 0.15829707082640831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31029355182, "tokens_per_second": 143201.93933459013, "data_wait_s": 1.4655000995844603e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 212100, "loss": 0.6130591575056314, "lr": 0.0003, "grad_norm": 0.15937313799624117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31044018740, "tokens_per_second": 149259.78657631867, "data_wait_s": 1.2718002835754305e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 212200, "loss": 0.6074743866920471, "lr": 0.0003, "grad_norm": 0.17296962229424495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31058690529, "tokens_per_second": 149128.21955117432, "data_wait_s": 1.2442997103789821e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 212300, "loss": 0.5837424118071795, "lr": 0.0003, "grad_norm": 0.14022896598437734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31073379662, "tokens_per_second": 148958.18816050663, "data_wait_s": 1.9877999875461683e-05, "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:e786a17ecffabfe168becd2a7fb89b8a644e843d192bfad73648a678df95b356
-size 366810
+oid sha256:70da0670fdf873c8105e24c3cb2940724a34ee8f7c1aace671eaf8321ccb3cb2
+size 368928