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": 76743.88558904106,
-      "required_pages_per_second_upper_estimate": 0.005309320075397733
+      "required_token_positions_per_second": 76504.96761960612,
+      "required_pages_per_second_upper_estimate": 0.005292791175906158
     },
     {
       "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": 9592.985698630133,
-      "required_pages_per_second_upper_estimate": 0.004016382787695411
+      "required_token_positions_per_second": 9563.120952450765,
+      "required_pages_per_second_upper_estimate": 0.004003879042116927
     },
     {
       "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": 19185.971397260266,
-      "required_pages_per_second_upper_estimate": 0.003995931873958687
+      "required_token_positions_per_second": 19126.24190490153,
+      "required_pages_per_second_upper_estimate": 0.003983491795873002
     },
     {
       "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": 28778.957095890397,
-      "required_pages_per_second_upper_estimate": 0.0017252545349755216
+      "required_token_positions_per_second": 28689.362857352295,
+      "required_pages_per_second_upper_estimate": 0.0017198834971776436
     },
     {
       "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": 28778.957095890397,
-      "required_pages_per_second_upper_estimate": 0.0022374373289511555
+      "required_token_positions_per_second": 28689.362857352295,
+      "required_pages_per_second_upper_estimate": 0.0022304717709882254
     },
     {
       "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": 28778.957095890397,
-      "required_pages_per_second_upper_estimate": 0.36914556118944597
+      "required_token_positions_per_second": 28689.362857352295,
+      "required_pages_per_second_upper_estimate": 0.36799634249627755
     }
   ],
   "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": 0.10490553395322877,
-    "producer_p95_seconds": 0.262934579001012,
-    "gpu_update_seconds": 0.9607019430161388,
-    "producer_batches_per_second": 9.532385588407962,
-    "gpu_batches_per_second": 1.0409055662576099,
-    "producer_headroom": 9.157781356362568,
+    "producer_batch_seconds": 0.09398608896086103,
+    "producer_p95_seconds": 0.272052423999412,
+    "gpu_update_seconds": 0.9637021267244552,
+    "producer_batches_per_second": 10.639872464705215,
+    "gpu_batches_per_second": 1.0376650339030777,
+    "producer_headroom": 10.253667722313386,
     "sustainable": true,
-    "buffer_seconds": 60.52422241001674,
+    "buffer_seconds": 60.71323398364068,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 274853.9888921495,
-    "upload_bytes_per_second": 3608555.4110745294,
+    "generation_bytes_per_second": 275883.36016573437,
+    "upload_bytes_per_second": 3585669.436596974,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.977267742856625,
+    "estimated_full_shards_per_hour": 59.198146855631094,
     "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": 1.5798416210000141,
-      "bytes": 6237294,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.5424033700001019,
       "bytes": 7705972,
@@ -562,6 +556,12 @@
       "bytes": 17299146,
       "pages": 27,
       "shards": 6
+    },
+    {
+      "seconds": 0.9336543670033279,
+      "bytes": 1145946,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -207,3 +207,4 @@
 {"step": 863200, "loss": 0.548261284828186, "lr": 3e-05, "grad_norm": 0.30352197081077575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126348016210, "tokens_per_second": 104951.52948531444, "data_wait_s": 2.045100336545147e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 863300, "loss": 0.53046053647995, "lr": 3e-05, "grad_norm": 0.3072887882862738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126362699196, "tokens_per_second": 152343.5660354792, "data_wait_s": 8.579001587349921e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 863400, "loss": 0.5254390239715576, "lr": 3e-05, "grad_norm": 0.30600288528215086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126377340544, "tokens_per_second": 151413.7889389153, "data_wait_s": 1.3708002370549366e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 863500, "loss": 0.5279927849769592, "lr": 3e-05, "grad_norm": 0.32024103289731815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126391995489, "tokens_per_second": 152755.33657753884, "data_wait_s": 1.255100141861476e-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 sha256:9658254e9e7e2852d1f4396833bbd479af58f9205b33f4a27e3f8afb1c6a0d06
-size 445959
+oid sha256:98b2794a38ed55cdf14996fff5b6e2ec6619cbce11088ae8124bf36e7e682d83
+size 448077