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": 77209.77279904924,
-      "required_pages_per_second_upper_estimate": 0.005341551233593361
+      "required_token_positions_per_second": 76301.24885801745,
+      "required_pages_per_second_upper_estimate": 0.005278697439286803
     },
     {
       "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": 9651.221599881155,
-      "required_pages_per_second_upper_estimate": 0.0040407649434453085
+      "required_token_positions_per_second": 9537.656107252182,
+      "required_pages_per_second_upper_estimate": 0.003993217443198718
     },
     {
       "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": 19302.44319976231,
-      "required_pages_per_second_upper_estimate": 0.004020189878851877
+      "required_token_positions_per_second": 19075.312214504364,
+      "required_pages_per_second_upper_estimate": 0.003972884484469528
     },
     {
       "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": 28953.664799643466,
-      "required_pages_per_second_upper_estimate": 0.0017357279950523006
+      "required_token_positions_per_second": 28612.968321756543,
+      "required_pages_per_second_upper_estimate": 0.00171530376141638
     },
     {
       "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": 28953.664799643466,
-      "required_pages_per_second_upper_estimate": 0.002251020084459981
+      "required_token_positions_per_second": 28612.968321756543,
+      "required_pages_per_second_upper_estimate": 0.0022245324318697058
     },
     {
       "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": 28953.664799643466,
-      "required_pages_per_second_upper_estimate": 0.37138652402667316
+      "required_token_positions_per_second": 28612.968321756543,
+      "required_pages_per_second_upper_estimate": 0.36701643541971685
     }
   ],
   "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": 0.10020453504682791,
-    "producer_p95_seconds": 0.3740277360047912,
-    "gpu_update_seconds": 0.9549050246772373,
-    "producer_batches_per_second": 9.979588244510857,
-    "gpu_batches_per_second": 1.0472245659593267,
-    "producer_headroom": 9.529558958893306,
+    "producer_batch_seconds": 0.15182135608580438,
+    "producer_p95_seconds": 0.35369676100526704,
+    "gpu_update_seconds": 0.966275141016292,
+    "producer_batches_per_second": 6.586688630516732,
+    "gpu_batches_per_second": 1.0349019213598287,
+    "producer_headroom": 6.364553485282962,
     "sustainable": true,
-    "buffer_seconds": 60.159016554665946,
+    "buffer_seconds": 60.875333884026396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344015.0653301798,
-    "upload_bytes_per_second": 5351231.68263034,
+    "generation_bytes_per_second": 310209.76005789376,
+    "upload_bytes_per_second": 5398279.594167308,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.8176247589974,
+    "estimated_full_shards_per_hour": 66.5637931947957,
     "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": 1.0037741340020148,
-      "bytes": 6334013,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.4788876900020114,
       "bytes": 12280922,
@@ -562,6 +556,12 @@
       "bytes": 886815,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.3266999849947751,
+      "bytes": 11836619,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -402,3 +402,4 @@
 {"step": 480400, "loss": 0.5549145340919495, "lr": 0.0003, "grad_norm": 0.14781810418308025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70315062178, "tokens_per_second": 150693.91314482232, "data_wait_s": 1.0480995115358382e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 480500, "loss": 0.5556878447532654, "lr": 0.0003, "grad_norm": 0.1685063295264039, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70329520877, "tokens_per_second": 150014.02147498846, "data_wait_s": 8.021997928153723e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 480600, "loss": 0.5902277827262878, "lr": 0.0003, "grad_norm": 0.15070081294110974, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70344147358, "tokens_per_second": 151402.92773309172, "data_wait_s": 1.3409000530373305e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 480700, "loss": 0.5943543314933777, "lr": 0.0003, "grad_norm": 0.1558133987893216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70358842762, "tokens_per_second": 152542.6546573282, "data_wait_s": 9.530005627311766e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5b8f8b867bf26d6d6811ee0cf99b7d86ad97b5e5ee8a4d4b779e49d6a32b9fa
-size 862101
+oid sha256:f6341f235583d18658d02c1fe2f53f8e6dc23aed2820ec51784c18b5985ac611
+size 864219