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": 76971.24271485917,
-      "required_pages_per_second_upper_estimate": 0.005325049168903043
+      "required_token_positions_per_second": 76786.25347748448,
+      "required_pages_per_second_upper_estimate": 0.0053122511842272945
     },
     {
       "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": 9621.405339357396,
-      "required_pages_per_second_upper_estimate": 0.004028281497798332
+      "required_token_positions_per_second": 9598.28168468556,
+      "required_pages_per_second_upper_estimate": 0.004018600106464054
     },
     {
       "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": 19242.81067871479,
-      "required_pages_per_second_upper_estimate": 0.004007769997333013
+      "required_token_positions_per_second": 19196.56336937112,
+      "required_pages_per_second_upper_estimate": 0.003998137902420289
     },
     {
       "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": 28864.216018072188,
-      "required_pages_per_second_upper_estimate": 0.0017303656771774836
+      "required_token_positions_per_second": 28794.845054056685,
+      "required_pages_per_second_upper_estimate": 0.0017262069938080832
     },
     {
       "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": 28864.216018072188,
-      "required_pages_per_second_upper_estimate": 0.002244065835136424
+      "required_token_positions_per_second": 28794.845054056685,
+      "required_pages_per_second_upper_estimate": 0.0022386725478148422
     },
     {
       "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": 28864.216018072188,
-      "required_pages_per_second_upper_estimate": 0.37023917109929566
+      "required_token_positions_per_second": 28794.845054056685,
+      "required_pages_per_second_upper_estimate": 0.36934935485764275
     }
   ],
   "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": 0.3996546975076285,
-    "producer_p95_seconds": 0.2843913569995493,
-    "gpu_update_seconds": 0.957864228243348,
-    "producer_batches_per_second": 2.502160005215283,
-    "gpu_batches_per_second": 1.0439892946351341,
-    "producer_headroom": 2.3967295623369087,
+    "producer_batch_seconds": 0.13876101382007278,
+    "producer_p95_seconds": 0.2885947600007057,
+    "gpu_update_seconds": 0.9601718622932783,
+    "producer_batches_per_second": 7.206635152555672,
+    "gpu_batches_per_second": 1.0414802175223048,
+    "producer_headroom": 6.919608295297585,
     "sustainable": true,
-    "buffer_seconds": 60.34544637933093,
+    "buffer_seconds": 60.490827324476534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309794.473975571,
-    "upload_bytes_per_second": 4898379.503515375,
+    "generation_bytes_per_second": 322149.4412047158,
+    "upload_bytes_per_second": 4784938.240880121,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.47468246889446,
+    "estimated_full_shards_per_hour": 69.12577082735163,
     "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": 1.6123849089999567,
-      "bytes": 16470150,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 1.350964933000796,
       "bytes": 1228633,
@@ -562,6 +556,12 @@
       "bytes": 6152144,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3146963539984426,
+      "bytes": 5075870,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -270,3 +270,4 @@
 {"step": 869500, "loss": 0.49472102522850037, "lr": 3e-05, "grad_norm": 0.299259477089786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127270536680, "tokens_per_second": 150904.69734010668, "data_wait_s": 9.16299904929474e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 869600, "loss": 0.5351412892341614, "lr": 3e-05, "grad_norm": 0.31434312061137293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127285195105, "tokens_per_second": 151602.47113576697, "data_wait_s": 1.963199974852614e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869700, "loss": 0.5524712800979614, "lr": 3e-05, "grad_norm": 0.3183505550061015, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127299888487, "tokens_per_second": 152071.72078037952, "data_wait_s": 1.468299888074398e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 869800, "loss": 0.5738760828971863, "lr": 3e-05, "grad_norm": 0.3299766884051443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127314599732, "tokens_per_second": 152909.14793350216, "data_wait_s": 1.777400029823184e-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:d900d7355f289cf8445de0cf25e396b8b6986d7414a57ce38232741f30e03e31
-size 580176
+oid sha256:5a433fc1b64cf31fdfba2888536153f63fdd0271a5f0dc7d56aac4d5942f659c
+size 582294