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": 77507.70753695644,
-      "required_pages_per_second_upper_estimate": 0.005362163050065615
+      "required_token_positions_per_second": 77416.7751981749,
+      "required_pages_per_second_upper_estimate": 0.005355872139876613
     },
     {
       "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": 9688.463442119555,
-      "required_pages_per_second_upper_estimate": 0.004056357325092425
+      "required_token_positions_per_second": 9677.096899771863,
+      "required_pages_per_second_upper_estimate": 0.004051598391171842
     },
     {
       "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": 19376.92688423911,
-      "required_pages_per_second_upper_estimate": 0.004035702866061544
+      "required_token_positions_per_second": 19354.193799543726,
+      "required_pages_per_second_upper_estimate": 0.004030968164031254
     },
     {
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.0017424257697838755
+      "required_token_positions_per_second": 29031.29069931559,
+      "required_pages_per_second_upper_estimate": 0.0017403815492097584
     },
     {
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.002259706252733432
+      "required_token_positions_per_second": 29031.29069931559,
+      "required_pages_per_second_upper_estimate": 0.0022570551567192404
     },
     {
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.3728196191218516
+      "required_token_positions_per_second": 29031.29069931559,
+      "required_pages_per_second_upper_estimate": 0.37238222571946983
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08243451003123425,
-    "producer_p95_seconds": 0.21277156500036654,
-    "gpu_update_seconds": 0.9512344300061482,
-    "producer_batches_per_second": 12.130841799400544,
-    "gpu_batches_per_second": 1.0512655644660978,
-    "producer_headroom": 11.539274384547534,
+    "producer_batch_seconds": 0.10638115483595811,
+    "producer_p95_seconds": 0.22856816000057734,
+    "gpu_update_seconds": 0.952351732699635,
+    "producer_batches_per_second": 9.400161161457781,
+    "gpu_batches_per_second": 1.0500322156870512,
+    "producer_headroom": 8.95225976977013,
     "sustainable": true,
-    "buffer_seconds": 60.87900352039348,
+    "buffer_seconds": 60.95051089277664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409692.50155643857,
-    "upload_bytes_per_second": 4323983.225698083,
+    "generation_bytes_per_second": 393433.2918256515,
+    "upload_bytes_per_second": 4260532.603409401,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.91047368068568,
+    "estimated_full_shards_per_hour": 84.42162576748999,
     "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": 1.8616032419995463,
-      "bytes": 11914182,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.137752229000398,
       "bytes": 7686898,
@@ -562,6 +556,12 @@
       "bytes": 6274938,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1267517289998068,
+      "bytes": 2389323,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 699800, "loss": 0.5427844524383545, "lr": 8e-05, "grad_norm": 0.2044449421545245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102431389470, "tokens_per_second": 151725.389850195, "data_wait_s": 1.3274999218992889e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 699900, "loss": 0.5266990065574646, "lr": 8e-05, "grad_norm": 0.2204417780876645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102446083649, "tokens_per_second": 153643.0955539443, "data_wait_s": 1.6406000213464722e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 700000, "loss": 0.5678468942642212, "lr": 8e-05, "grad_norm": 0.20413006877333337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102460742219, "tokens_per_second": 152822.87951846494, "data_wait_s": 1.8226999600301497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 700100, "loss": 0.5273600816726685, "lr": 8e-05, "grad_norm": 0.20572507000430548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 102475568774, "tokens_per_second": 145293.57576594577, "data_wait_s": 1.4528000974678434e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ad733545188eaa1840f0429b7078dded03b419b6a638ba60c96ceac6058e509
-size 320214
+oid sha256:604810906fd914bb5c766f87960ebeb74e068c0e056d52dc146d8b0c8608c554
+size 322332