Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76792.93497126369,
-      "required_pages_per_second_upper_estimate": 0.005312713425470142
+      "required_token_positions_per_second": 76769.44490308352,
+      "required_pages_per_second_upper_estimate": 0.00531108832804897
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9599.116871407961,
-      "required_pages_per_second_upper_estimate": 0.004018949781704041
+      "required_token_positions_per_second": 9596.18061288544,
+      "required_pages_per_second_upper_estimate": 0.0040177204315767125
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19198.233742815923,
-      "required_pages_per_second_upper_estimate": 0.003998485797158132
+      "required_token_positions_per_second": 19192.36122577088,
+      "required_pages_per_second_upper_estimate": 0.003997262706726339
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28797.350614223888,
-      "required_pages_per_second_upper_estimate": 0.001726357198314338
+      "required_token_positions_per_second": 28788.541838656325,
+      "required_pages_per_second_upper_estimate": 0.0017258291256692843
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28797.350614223888,
-      "required_pages_per_second_upper_estimate": 0.0022388673440970476
+      "required_token_positions_per_second": 28788.541838656325,
+      "required_pages_per_second_upper_estimate": 0.0022381825005423784
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28797.350614223888,
-      "required_pages_per_second_upper_estimate": 0.3693814934932067
+      "required_token_positions_per_second": 28788.541838656325,
+      "required_pages_per_second_upper_estimate": 0.3692685039783523
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1001761312888334,
-    "producer_p95_seconds": 0.3633046850009123,
-    "gpu_update_seconds": 0.9600883209840774,
-    "producer_batches_per_second": 9.98241783880378,
-    "gpu_batches_per_second": 1.0415708410815931,
-    "producer_headroom": 9.584002782218622,
+    "producer_batch_seconds": 0.2893251365311471,
+    "producer_p95_seconds": 0.3644270360018709,
+    "gpu_update_seconds": 0.9603820907273309,
+    "producer_batches_per_second": 3.4563191155440642,
+    "gpu_batches_per_second": 1.0412522366412154,
+    "producer_headroom": 3.3193869784070476,
     "sustainable": true,
-    "buffer_seconds": 60.485564221996874,
+    "buffer_seconds": 60.50407171582185,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314549.9880951152,
-    "upload_bytes_per_second": 3684831.429626612,
+    "generation_bytes_per_second": 353563.1825064903,
+    "upload_bytes_per_second": 3706747.296253358,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.49510509624568,
+    "estimated_full_shards_per_hour": 75.86642843624146,
     "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": 4.01536119299999,
-      "bytes": 5846921,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 2.7826384430009057,
       "bytes": 7726579,
@@ -562,6 +556,12 @@
       "bytes": 6207448,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 2.0409108199964976,
+      "bytes": 1044098,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -400,3 +400,4 @@
 {"step": 882500, "loss": 0.5453665256500244, "lr": 3e-05, "grad_norm": 0.32312718729762313, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129173267711, "tokens_per_second": 150456.0791782488, "data_wait_s": 1.6260994016192853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882600, "loss": 0.5886988639831543, "lr": 3e-05, "grad_norm": 0.36577334299302855, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129187683497, "tokens_per_second": 149640.32573357943, "data_wait_s": 1.8963997717946768e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 882700, "loss": 0.5182936787605286, "lr": 3e-05, "grad_norm": 0.3073328887564094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129202265162, "tokens_per_second": 151357.08025770652, "data_wait_s": 1.7841004591900855e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 882800, "loss": 0.5007718801498413, "lr": 3e-05, "grad_norm": 0.29883861408068824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 129216915075, "tokens_per_second": 148817.6257671427, "data_wait_s": 1.154300116468221e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:62cdf7cc4ff7f9c9ded670727f7bdf8dbe134298455223aa913e95609601d205
-size 857865
+oid sha256:7b1d4de54d8993d6ab45993ab0aa60f43d0c0b6fa1f40ddf7345c40fe1586d83
+size 859983