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": 77429.51199844645,
-      "required_pages_per_second_upper_estimate": 0.0053567533012728475
+      "required_token_positions_per_second": 77515.81336660638,
+      "required_pages_per_second_upper_estimate": 0.005362723830168914
     },
     {
       "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": 9678.688999805807,
-      "required_pages_per_second_upper_estimate": 0.004052264970209225
+      "required_token_positions_per_second": 9689.476670825798,
+      "required_pages_per_second_upper_estimate": 0.004056781542793108
     },
     {
       "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": 19357.377999611614,
-      "required_pages_per_second_upper_estimate": 0.0040316313489323925
+      "required_token_positions_per_second": 19378.953341651595,
+      "required_pages_per_second_upper_estimate": 0.004036124923699288
     },
     {
       "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": 29036.06699941742,
-      "required_pages_per_second_upper_estimate": 0.0017406678811078755
+      "required_token_positions_per_second": 29068.430012477394,
+      "required_pages_per_second_upper_estimate": 0.0017426079943253612
     },
     {
       "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": 29036.06699941742,
-      "required_pages_per_second_upper_estimate": 0.002257426493043434
+      "required_token_positions_per_second": 29068.430012477394,
+      "required_pages_per_second_upper_estimate": 0.0022599425749589967
     },
     {
       "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": 29036.06699941742,
-      "required_pages_per_second_upper_estimate": 0.37244349096878465
+      "required_token_positions_per_second": 29068.430012477394,
+      "required_pages_per_second_upper_estimate": 0.3728586089516219
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.38948444139839467,
+    "producer_batch_seconds": 0.0747855775624231,
     "producer_p95_seconds": 0.20644337199883012,
-    "gpu_update_seconds": 0.9521950751992247,
-    "producer_batches_per_second": 2.5674966538063146,
-    "gpu_batches_per_second": 1.0502049695969844,
-    "producer_headroom": 2.444757669344862,
+    "gpu_update_seconds": 0.9511349594089641,
+    "producer_batches_per_second": 13.371562172737192,
+    "gpu_batches_per_second": 1.0513755068170354,
+    "producer_headroom": 12.718160244400828,
     "sustainable": true,
-    "buffer_seconds": 60.940484812750384,
+    "buffer_seconds": 60.8726374021737,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370440.62564360816,
-    "upload_bytes_per_second": 5626184.462655927,
+    "generation_bytes_per_second": 369808.39453087514,
+    "upload_bytes_per_second": 5415588.325594185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.48793484669861,
+    "estimated_full_shards_per_hour": 79.35227276749316,
     "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": 1.2055649489993812,
-      "bytes": 6199136,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1954915420010366,
       "bytes": 12042262,
@@ -562,6 +556,12 @@
       "bytes": 10017513,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 4.1799396490005165,
+      "bytes": 5932760,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -158,3 +158,4 @@
 {"step": 572900, "loss": 0.5719613432884216, "lr": 0.0003, "grad_norm": 0.15580971464201138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83855622847, "tokens_per_second": 152768.45641343156, "data_wait_s": 9.494999176240526e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 573000, "loss": 0.5801893472671509, "lr": 0.0003, "grad_norm": 0.1543706092311972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83870326717, "tokens_per_second": 154341.376915835, "data_wait_s": 1.0422001651022583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 573100, "loss": 0.5934357643127441, "lr": 0.0003, "grad_norm": 0.1694824302473004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83885041144, "tokens_per_second": 144186.33631227972, "data_wait_s": 1.0832998668774962e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 573200, "loss": 0.5817840695381165, "lr": 0.0003, "grad_norm": 0.17621079851627106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83899574908, "tokens_per_second": 151516.6430221442, "data_wait_s": 1.6834999769343995e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a49aca90d29f7302e969365fc2379fa8853610748cb7c490acc4b1620342f42d
-size 341394
+oid sha256:fe6fc0f2e86fb179f432c475eeb77b624dee3ed35dd17cfaa58fd546d84f7115
+size 343512