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": 76930.23044474196,
-      "required_pages_per_second_upper_estimate": 0.005322211844894757
+      "required_token_positions_per_second": 77389.55419018122,
+      "required_pages_per_second_upper_estimate": 0.005353988927382171
     },
     {
       "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": 9616.278805592745,
-      "required_pages_per_second_upper_estimate": 0.0040261351251652715
+      "required_token_positions_per_second": 9673.694273772653,
+      "required_pages_per_second_upper_estimate": 0.0040501737827208836
     },
     {
       "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": 19232.55761118549,
-      "required_pages_per_second_upper_estimate": 0.004005634553758233
+      "required_token_positions_per_second": 19347.388547545306,
+      "required_pages_per_second_upper_estimate": 0.004029550809506547
     },
     {
       "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": 28848.836416778235,
-      "required_pages_per_second_upper_estimate": 0.0017294436935632027
+      "required_token_positions_per_second": 29021.08282131796,
+      "required_pages_per_second_upper_estimate": 0.0017397696025103278
     },
     {
       "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": 28848.836416778235,
-      "required_pages_per_second_upper_estimate": 0.002242870138783536
+      "required_token_positions_per_second": 29021.08282131796,
+      "required_pages_per_second_upper_estimate": 0.002256261539104635
     },
     {
       "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": 28848.836416778235,
-      "required_pages_per_second_upper_estimate": 0.3700418980872261
+      "required_token_positions_per_second": 29021.08282131796,
+      "required_pages_per_second_upper_estimate": 0.3722512900208817
     }
   ],
   "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": 0.10528680584376104,
-    "producer_p95_seconds": 0.22316760000103386,
-    "gpu_update_seconds": 0.9583748751793734,
-    "producer_batches_per_second": 9.497866251958833,
-    "gpu_batches_per_second": 1.0434330301207404,
-    "producer_headroom": 9.102516383691428,
+    "producer_batch_seconds": 0.32744307142189655,
+    "producer_p95_seconds": 0.2328790610008582,
+    "gpu_update_seconds": 0.9526867129744264,
+    "producer_batches_per_second": 3.0539659784449746,
+    "gpu_batches_per_second": 1.0496630071367896,
+    "producer_headroom": 2.9094728095404707,
     "sustainable": true,
-    "buffer_seconds": 60.37761713630052,
+    "buffer_seconds": 60.019262917388865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454900.8052649066,
-    "upload_bytes_per_second": 4625336.309495164,
+    "generation_bytes_per_second": 436699.35306817037,
+    "upload_bytes_per_second": 3421882.600967228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.61112326107406,
+    "estimated_full_shards_per_hour": 93.70551532777627,
     "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": 1.6813211230000888,
-      "bytes": 7642707,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.8502533870005209,
       "bytes": 10875382,
@@ -562,6 +556,12 @@
       "bytes": 6235687,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 33.88602663399979,
+      "bytes": 7355362,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -145,3 +145,4 @@
 {"step": 857000, "loss": 0.488033652305603, "lr": 3e-05, "grad_norm": 0.3102570772317539, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125441091982, "tokens_per_second": 153238.26787279305, "data_wait_s": 1.3568998838309199e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857100, "loss": 0.5046170949935913, "lr": 3e-05, "grad_norm": 0.3271241915699481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125455590368, "tokens_per_second": 150637.4781559886, "data_wait_s": 1.858899850049056e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857200, "loss": 0.4982972741127014, "lr": 3e-05, "grad_norm": 0.31009411213021565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125470173330, "tokens_per_second": 151760.4849812411, "data_wait_s": 1.2570999388117343e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 857300, "loss": 0.5432097911834717, "lr": 3e-05, "grad_norm": 0.30741004342921235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125484708242, "tokens_per_second": 151424.46058662882, "data_wait_s": 1.3279999620863236e-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:d1ee823080bff41cf406599dd1b6479040a88be504680ce71be8d2ab3714cd2a
-size 313860
+oid sha256:4bac10541f9b60e97389dda732f5982603c3431bff366c56b90d29a0fa4ff8f9
+size 315978