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": 76693.90823857264,
-      "required_pages_per_second_upper_estimate": 0.005305862526328898
+      "required_token_positions_per_second": 76168.83426116995,
+      "required_pages_per_second_upper_estimate": 0.005269536690232698
     },
     {
       "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": 9586.73852982158,
-      "required_pages_per_second_upper_estimate": 0.004013767228570989
+      "required_token_positions_per_second": 9521.104282646244,
+      "required_pages_per_second_upper_estimate": 0.003986287539877613
     },
     {
       "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": 19173.47705964316,
-      "required_pages_per_second_upper_estimate": 0.003993329632930884
+      "required_token_positions_per_second": 19042.208565292487,
+      "required_pages_per_second_upper_estimate": 0.00396598986734063
     },
     {
       "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": 28760.215589464737,
-      "required_pages_per_second_upper_estimate": 0.0017241310102819266
+      "required_token_positions_per_second": 28563.31284793873,
+      "required_pages_per_second_upper_estimate": 0.0017123269915805205
     },
     {
       "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": 28760.215589464737,
-      "required_pages_per_second_upper_estimate": 0.002235980259261735
+      "required_token_positions_per_second": 28563.31284793873,
+      "required_pages_per_second_upper_estimate": 0.002220671937191718
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28760.215589464737,
-      "required_pages_per_second_upper_estimate": 0.36890516526807937
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28563.31284793873,
+      "required_pages_per_second_upper_estimate": 0.3663795083174758
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05656698913284686,
-    "producer_p95_seconds": 0.18373191299997416,
-    "gpu_update_seconds": 0.9613279814956548,
-    "producer_batches_per_second": 17.6781549686428,
-    "gpu_batches_per_second": 1.0402277050587652,
-    "producer_headroom": 16.994505032572764,
+    "producer_batch_seconds": 0.37420466701558297,
+    "producer_p95_seconds": 0.21539342300002318,
+    "gpu_update_seconds": 0.9679549479147765,
+    "producer_batches_per_second": 2.672334388492159,
+    "gpu_batches_per_second": 1.033105933446858,
+    "producer_headroom": 2.586699293823794,
     "sustainable": true,
-    "buffer_seconds": 60.56366283422625,
+    "buffer_seconds": 60.013206770716145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364853.27285089326,
-    "upload_bytes_per_second": 5302397.638489437,
+    "generation_bytes_per_second": 333105.4027836271,
+    "upload_bytes_per_second": 5407922.942715552,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.2890190042982,
+    "estimated_full_shards_per_hour": 71.47666514045343,
     "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": [
@@ -376,6 +376,12 @@
       "bytes": 7715241,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1387970109999515,
+      "bytes": 10453428,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 767500, "loss": 0.529920220375061, "lr": 3e-05, "grad_norm": 0.24821814876417622, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112339415179, "tokens_per_second": 153615.94682883014, "data_wait_s": 1.080899983207928e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 767600, "loss": 0.5293038487434387, "lr": 3e-05, "grad_norm": 0.24808115644714873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112354162086, "tokens_per_second": 153498.4234518902, "data_wait_s": 1.395699928252725e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 767700, "loss": 0.4996536374092102, "lr": 3e-05, "grad_norm": 0.24608894016068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112368992756, "tokens_per_second": 153523.3098604039, "data_wait_s": 1.1235000783926807e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 767800, "loss": 0.5399699807167053, "lr": 3e-05, "grad_norm": 0.2565081637582412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112383712276, "tokens_per_second": 152727.7079886464, "data_wait_s": 1.0433000170451123e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6cd2b79061cf91b040519e2fa060437d5fa9773f25ae2b7270f2c2060085845
-size 164034
+oid sha256:b91113f4dac79fa1f21546b3bd2acfdebb93a490fe92998ec444afa1e9361f4c
+size 166152