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": 77300.98167934327,
-      "required_pages_per_second_upper_estimate": 0.005347861275565861
+      "required_token_positions_per_second": 77380.05617286645,
+      "required_pages_per_second_upper_estimate": 0.005353331832505898
     },
     {
       "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": 9662.622709917909,
-      "required_pages_per_second_upper_estimate": 0.004045538350135446
+      "required_token_positions_per_second": 9672.507021608306,
+      "required_pages_per_second_upper_estimate": 0.004049676705032316
     },
     {
       "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": 19325.245419835817,
-      "required_pages_per_second_upper_estimate": 0.004024938979958207
+      "required_token_positions_per_second": 19345.01404321661,
+      "required_pages_per_second_upper_estimate": 0.00402905626287477
     },
     {
       "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": 28987.868129753726,
-      "required_pages_per_second_upper_estimate": 0.0017377784324669496
+      "required_token_positions_per_second": 29017.52106482492,
+      "required_pages_per_second_upper_estimate": 0.0017395560806470573
     },
     {
       "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": 28987.868129753726,
-      "required_pages_per_second_upper_estimate": 0.002253679242931504
+      "required_token_positions_per_second": 29017.52106482492,
+      "required_pages_per_second_upper_estimate": 0.0022559846282038123
     },
     {
       "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": 28987.868129753726,
-      "required_pages_per_second_upper_estimate": 0.37182524762065294
+      "required_token_positions_per_second": 29017.52106482492,
+      "required_pages_per_second_upper_estimate": 0.37220560363290517
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.2756291726641322,
-    "producer_p95_seconds": 0.16944977099956304,
-    "gpu_update_seconds": 0.9537783143018214,
-    "producer_batches_per_second": 3.628062988885975,
-    "gpu_batches_per_second": 1.0484616655726897,
-    "producer_headroom": 3.460367801720493,
+    "producer_batch_seconds": 0.0951179124296786,
+    "producer_p95_seconds": 0.16730843099958292,
+    "gpu_update_seconds": 0.9528036505335717,
+    "producer_batches_per_second": 10.51326689638303,
+    "gpu_batches_per_second": 1.0495341820321513,
+    "producer_headroom": 10.017079077907505,
     "sustainable": true,
-    "buffer_seconds": 60.088033801014745,
+    "buffer_seconds": 60.02662998361502,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297692.3094708694,
-    "upload_bytes_per_second": 5922209.00531333,
+    "generation_bytes_per_second": 299636.8965522193,
+    "upload_bytes_per_second": 5843749.448945796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.87783969015658,
+    "estimated_full_shards_per_hour": 64.29510281014379,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 11842449,
       "pages": 27,
       "shards": 3
+    },
+    {
+      "seconds": 1.2498191829999996,
+      "bytes": 5414761,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -29,3 +29,4 @@
 {"step": 487900, "loss": 0.5991864204406738, "lr": 0.0003, "grad_norm": 0.1773661234737318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71412514648, "tokens_per_second": 136685.44896800318, "data_wait_s": 1.4724000720889308e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488000, "loss": 0.5991197228431702, "lr": 0.0003, "grad_norm": 0.15802435942714238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71427219379, "tokens_per_second": 153058.96275634225, "data_wait_s": 1.2954999874636997e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 488100, "loss": 0.6014329195022583, "lr": 0.0003, "grad_norm": 0.20564692473549886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71441950975, "tokens_per_second": 154479.18494264406, "data_wait_s": 9.575000149197876e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 488200, "loss": 0.5942641496658325, "lr": 0.0003, "grad_norm": 0.16873086263281123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71456706571, "tokens_per_second": 153355.76433489085, "data_wait_s": 1.0932000805041753e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39493cf93f69ecf508bd7de6b03ab3206fd6bb49ba021b4df5624694439f4d9a
-size 65823
+oid sha256:e42b6f8b15e61de5b43650ea2c45618a7e8c9979d7ea7e078a2220421b0f3228
+size 67941