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": 77401.84559215908,
-      "required_pages_per_second_upper_estimate": 0.0053548392750909805
+      "required_token_positions_per_second": 76750.1843071453,
+      "required_pages_per_second_upper_estimate": 0.0053097558353312245
     },
     {
       "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": 9675.230699019885,
-      "required_pages_per_second_upper_estimate": 0.004050817051887692
+      "required_token_positions_per_second": 9593.773038393163,
+      "required_pages_per_second_upper_estimate": 0.004016712430412668
     },
     {
       "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": 19350.46139803977,
-      "required_pages_per_second_upper_estimate": 0.004030190803228027
+      "required_token_positions_per_second": 19187.546076786326,
+      "required_pages_per_second_upper_estimate": 0.003996259838176874
     },
     {
       "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": 29025.692097059655,
-      "required_pages_per_second_upper_estimate": 0.001740045921553086
+      "required_token_positions_per_second": 28781.319115179485,
+      "required_pages_per_second_upper_estimate": 0.0017253961344253078
     },
     {
       "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": 29025.692097059655,
-      "required_pages_per_second_upper_estimate": 0.0022566198900194905
+      "required_token_positions_per_second": 28781.319115179485,
+      "required_pages_per_second_upper_estimate": 0.0022376209655614572
     },
     {
       "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": 29025.692097059655,
-      "required_pages_per_second_upper_estimate": 0.3723104128610415
+      "required_token_positions_per_second": 28781.319115179485,
+      "required_pages_per_second_upper_estimate": 0.3691758586367477
     }
   ],
   "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": 0.3847329208984718,
-    "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,
+    "producer_batch_seconds": 0.08349064867183387,
+    "producer_p95_seconds": 0.18002966200037918,
+    "gpu_update_seconds": 0.9606231003296244,
+    "producer_batches_per_second": 11.977389275421412,
+    "gpu_batches_per_second": 1.0409909980895358,
+    "producer_headroom": 11.505756819610111,
     "sustainable": true,
-    "buffer_seconds": 60.00973186704648,
+    "buffer_seconds": 60.519255320766334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329606.36675492726,
-    "upload_bytes_per_second": 4649249.339470346,
+    "generation_bytes_per_second": 316766.254538016,
+    "upload_bytes_per_second": 4787377.77284189,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.72585346208442,
+    "estimated_full_shards_per_hour": 67.97066428284988,
     "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": 1.576328019000016,
-      "bytes": 582303,
-      "pages": 17,
-      "shards": 1
-    },
     {
       "seconds": 1.4797563120000632,
       "bytes": 958250,
@@ -562,6 +556,12 @@
       "bytes": 5040157,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2845682920014951,
+      "bytes": 11638802,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -78,3 +78,4 @@
 {"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}
+{"step": 850600, "loss": 0.5145455598831177, "lr": 3e-05, "grad_norm": 0.3018728292464876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124504502492, "tokens_per_second": 154343.01561946663, "data_wait_s": 1.0376999853178859e-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 sha256:6acbf2075da9957172f70dd193dc87bc4335085e6d73e3d84c9ebecea8b8c2c5
-size 171171
+oid sha256:ef23cb695d6cfb0924e0824f633fd57fbef752622e464b0e1d909000f04ebb3d
+size 173289