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": 74811.34385201805,
-      "required_pages_per_second_upper_estimate": 0.005175622353915861
+      "required_token_positions_per_second": 74954.80635385367,
+      "required_pages_per_second_upper_estimate": 0.005185547422671715
     },
     {
       "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": 9351.417981502256,
-      "required_pages_per_second_upper_estimate": 0.003915243429041533
+      "required_token_positions_per_second": 9369.350794231708,
+      "required_pages_per_second_upper_estimate": 0.003922751523251641
     },
     {
       "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": 18702.835963004512,
-      "required_pages_per_second_upper_estimate": 0.0038953075041414193
+      "required_token_positions_per_second": 18738.701588463417,
+      "required_pages_per_second_upper_estimate": 0.003902777368084361
     },
     {
       "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": 28054.25394450677,
-      "required_pages_per_second_upper_estimate": 0.0016818096876077136
+      "required_token_positions_per_second": 28108.052382695125,
+      "required_pages_per_second_upper_estimate": 0.0016850348218316468
     },
     {
       "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": 28054.25394450677,
-      "required_pages_per_second_upper_estimate": 0.002181094846563362
+      "required_token_positions_per_second": 28108.052382695125,
+      "required_pages_per_second_upper_estimate": 0.0021852774384981855
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28054.25394450677,
-      "required_pages_per_second_upper_estimate": 0.35984984728911595
+      "required_token_positions_per_second": 28108.052382695125,
+      "required_pages_per_second_upper_estimate": 0.3605399158899337
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4726736822422737,
-    "producer_p95_seconds": 0.2660830020013236,
-    "gpu_update_seconds": 0.9855189895510903,
-    "producer_batches_per_second": 2.115624452066362,
-    "gpu_batches_per_second": 1.0146937913956442,
-    "producer_headroom": 2.0849880722700203,
+    "producer_batch_seconds": 0.30287808456249365,
+    "producer_p95_seconds": 0.23224182299964014,
+    "gpu_update_seconds": 0.9836327193207325,
+    "producer_batches_per_second": 3.301658492209816,
+    "gpu_batches_per_second": 1.0166396261102113,
+    "producer_headroom": 3.247619320960731,
     "sustainable": true,
-    "buffer_seconds": 60.116658362616505,
+    "buffer_seconds": 60.00159587856468,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348060.4058647902,
-    "upload_bytes_per_second": 4785470.974881097,
+    "generation_bytes_per_second": 315428.56909322215,
+    "upload_bytes_per_second": 4882857.784431501,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.68566066701679,
+    "estimated_full_shards_per_hour": 67.68362812612055,
     "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": 1.0668450510002003,
-      "bytes": 1106491,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.397956458000408,
       "bytes": 11076576,
@@ -562,6 +556,12 @@
       "bytes": 1094803,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.8326161909990333,
+      "bytes": 13393499,
+      "pages": 34,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -140,3 +140,4 @@
 {"step": 734000, "loss": 0.5088242292404175, "lr": 5e-05, "grad_norm": 0.22126292743866083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107437593647, "tokens_per_second": 148952.50170638604, "data_wait_s": 2.1376999939093366e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 734100, "loss": 0.5477070212364197, "lr": 5e-05, "grad_norm": 0.22360780055394486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107452102670, "tokens_per_second": 148106.61874349456, "data_wait_s": 1.3844999557477422e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 734200, "loss": 0.4480799734592438, "lr": 5e-05, "grad_norm": 0.27367567516997054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107466768097, "tokens_per_second": 150377.50924507514, "data_wait_s": 2.246999974886421e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 734300, "loss": 0.5217543840408325, "lr": 5e-05, "grad_norm": 0.2353259790457653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107481445030, "tokens_per_second": 150577.21805593575, "data_wait_s": 2.1955000192974694e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f0302af1a3cb158736969121e3b94084ce06a9886762b2e35ef8e61009f4a7b
-size 302487
+oid sha256:93e2717071dd953c1ded02e0e9e3cfb9e24eede4f4f4175ff120888bd42bf612
+size 304605