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": 76901.01235069444,
-      "required_pages_per_second_upper_estimate": 0.005320190469353231
+      "required_token_positions_per_second": 77401.84559215908,
+      "required_pages_per_second_upper_estimate": 0.0053548392750909805
     },
     {
       "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": 9612.626543836805,
-      "required_pages_per_second_upper_estimate": 0.004024605999435957
+      "required_token_positions_per_second": 9675.230699019885,
+      "required_pages_per_second_upper_estimate": 0.004050817051887692
     },
     {
       "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": 19225.25308767361,
-      "required_pages_per_second_upper_estimate": 0.0040041132141439475
+      "required_token_positions_per_second": 19350.46139803977,
+      "required_pages_per_second_upper_estimate": 0.004030190803228027
     },
     {
       "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": 28837.879631510412,
-      "required_pages_per_second_upper_estimate": 0.0017287868510164391
+      "required_token_positions_per_second": 29025.692097059655,
+      "required_pages_per_second_upper_estimate": 0.001740045921553086
     },
     {
       "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": 28837.879631510412,
-      "required_pages_per_second_upper_estimate": 0.0022420182969227678
+      "required_token_positions_per_second": 29025.692097059655,
+      "required_pages_per_second_upper_estimate": 0.0022566198900194905
     },
     {
       "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": 28837.879631510412,
-      "required_pages_per_second_upper_estimate": 0.3699013562102899
+      "required_token_positions_per_second": 29025.692097059655,
+      "required_pages_per_second_upper_estimate": 0.3723104128610415
     }
   ],
   "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": 0.10823759293752744,
-    "producer_p95_seconds": 0.2043575800007602,
-    "gpu_update_seconds": 0.9587390041599917,
-    "producer_batches_per_second": 9.238934208165363,
-    "gpu_batches_per_second": 1.0430367343572922,
-    "producer_headroom": 8.857726582236142,
+    "producer_batch_seconds": 0.3847329208984718,
+    "producer_p95_seconds": 0.1850976530004118,
+    "gpu_update_seconds": 0.9525354264610553,
+    "producer_batches_per_second": 2.599205697460688,
+    "gpu_batches_per_second": 1.049829719945734,
+    "producer_headroom": 2.475835507490721,
     "sustainable": true,
-    "buffer_seconds": 60.400557262079474,
+    "buffer_seconds": 60.00973186704648,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368501.81945720577,
-    "upload_bytes_per_second": 4585185.85955862,
+    "generation_bytes_per_second": 329606.36675492726,
+    "upload_bytes_per_second": 4649249.339470346,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.07191217219477,
+    "estimated_full_shards_per_hour": 70.72585346208442,
     "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": 1.3177459699999758,
-      "bytes": 94124,
-      "pages": 4,
-      "shards": 1
-    },
     {
       "seconds": 1.576328019000016,
       "bytes": 582303,
@@ -562,6 +556,12 @@
       "bytes": 7396091,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1367835250002827,
+      "bytes": 5040157,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -77,3 +77,4 @@
 {"step": 850200, "loss": 0.5152894854545593, "lr": 3e-05, "grad_norm": 0.2985612132745051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124446006405, "tokens_per_second": 151491.3248043859, "data_wait_s": 1.0863000170502346e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 850300, "loss": 0.5646606683731079, "lr": 3e-05, "grad_norm": 0.31378122123391483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124460470927, "tokens_per_second": 150344.83567860647, "data_wait_s": 2.0165000023553148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 850400, "loss": 0.5389873385429382, "lr": 3e-05, "grad_norm": 0.29445258264390717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124474996158, "tokens_per_second": 139417.352217659, "data_wait_s": 1.1802000699390192e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 850500, "loss": 0.4876767694950104, "lr": 3e-05, "grad_norm": 0.28623268586402095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124489747449, "tokens_per_second": 152333.05123555998, "data_wait_s": 9.347000741399825e-06, "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 sha256:11ba6fa753e7d341cf87cbec7538ff1cfdd8f4755b89f11bbc6583a2f39de02b
-size 169053
+oid sha256:6acbf2075da9957172f70dd193dc87bc4335085e6d73e3d84c9ebecea8b8c2c5
+size 171171