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": 77223.1760658397,
-      "required_pages_per_second_upper_estimate": 0.005342478502689788
+      "required_token_positions_per_second": 77312.01773995301,
+      "required_pages_per_second_upper_estimate": 0.005348624775846031
     },
     {
       "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": 9652.897008229962,
-      "required_pages_per_second_upper_estimate": 0.004041466401934449
+      "required_token_positions_per_second": 9664.002217494126,
+      "required_pages_per_second_upper_estimate": 0.004046115921150215
     },
     {
       "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": 19305.794016459924,
-      "required_pages_per_second_upper_estimate": 0.0040208877656030114
+      "required_token_positions_per_second": 19328.004434988252,
+      "required_pages_per_second_upper_estimate": 0.004025513610054337
     },
     {
       "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": 28958.691024689884,
-      "required_pages_per_second_upper_estimate": 0.001736029309569232
+      "required_token_positions_per_second": 28992.006652482378,
+      "required_pages_per_second_upper_estimate": 0.001738026530585374
     },
     {
       "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": 28958.691024689884,
-      "required_pages_per_second_upper_estimate": 0.0022514108513493124
+      "required_token_positions_per_second": 28992.006652482378,
+      "required_pages_per_second_upper_estimate": 0.002254000994870219
     },
     {
       "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": 28958.691024689884,
-      "required_pages_per_second_upper_estimate": 0.37145099504482865
+      "required_token_positions_per_second": 28992.006652482378,
+      "required_pages_per_second_upper_estimate": 0.37187833214661664
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.3648615481170623,
-    "producer_p95_seconds": 0.28322975899936864,
-    "gpu_update_seconds": 0.9547392862622001,
-    "producer_batches_per_second": 2.740765655248384,
-    "gpu_batches_per_second": 1.0474063593999525,
-    "producer_headroom": 2.6167166455037933,
+    "producer_batch_seconds": 0.07733573187499587,
+    "producer_p95_seconds": 0.27561638900078833,
+    "gpu_update_seconds": 0.9536421652839507,
+    "producer_batches_per_second": 12.9306334310818,
+    "gpu_batches_per_second": 1.0486113517246232,
+    "producer_headroom": 12.331197263709889,
     "sustainable": true,
-    "buffer_seconds": 60.1485750345186,
+    "buffer_seconds": 60.079456412888895,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296809.19681568356,
-    "upload_bytes_per_second": 6157459.921528537,
+    "generation_bytes_per_second": 295298.9474612483,
+    "upload_bytes_per_second": 6197965.677843467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.68834427216416,
+    "estimated_full_shards_per_hour": 63.364279917508,
     "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": 1.116981245999341,
-      "bytes": 6115742,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.009036264000315,
       "bytes": 6506928,
@@ -562,6 +556,12 @@
       "bytes": 8511115,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2614193410008738,
+      "bytes": 9853756,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -255,3 +255,4 @@
 {"step": 510500, "loss": 0.6076563596725464, "lr": 0.0003, "grad_norm": 0.15903708875644673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74719951999, "tokens_per_second": 153463.67566932138, "data_wait_s": 1.971500023500994e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510600, "loss": 0.5998474955558777, "lr": 0.0003, "grad_norm": 0.16425945716733695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74734713757, "tokens_per_second": 149271.93040092214, "data_wait_s": 1.3688000763067976e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510700, "loss": 0.5512087345123291, "lr": 0.0003, "grad_norm": 0.17483065358041508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74749215150, "tokens_per_second": 150938.28669991807, "data_wait_s": 1.5644000086467713e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 510800, "loss": 0.5592138767242432, "lr": 0.0003, "grad_norm": 0.15223429740935868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74763728380, "tokens_per_second": 150930.57722491908, "data_wait_s": 1.7693000700091943e-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 sha256:1382a375dfd44d223906c1bb980818f7b4114d2977ca9374ac6e8284d179f234
-size 548406
+oid sha256:d5f7d25bf4deb5b3691cfaef222521d97c34d26b73145aa63e2e0db557542517
+size 550524