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": 76461.51968750442,
-      "required_pages_per_second_upper_estimate": 0.005289785347150271
+      "required_token_positions_per_second": 75780.09635265477,
+      "required_pages_per_second_upper_estimate": 0.0052426428997254
     },
     {
       "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": 9557.689960938053,
-      "required_pages_per_second_upper_estimate": 0.004001605199382554
+      "required_token_positions_per_second": 9472.512044081846,
+      "required_pages_per_second_upper_estimate": 0.0039659429843120295
     },
     {
       "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": 19115.379921876105,
-      "required_pages_per_second_upper_estimate": 0.003981229531258561
+      "required_token_positions_per_second": 18945.024088163693,
+      "required_pages_per_second_upper_estimate": 0.003945748903681714
     },
     {
       "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": 28673.069882814158,
-      "required_pages_per_second_upper_estimate": 0.001718906758232014
+      "required_token_positions_per_second": 28417.53613224554,
+      "required_pages_per_second_upper_estimate": 0.0017035879000628703
     },
     {
       "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": 28673.069882814158,
-      "required_pages_per_second_upper_estimate": 0.0022292050638831064
+      "required_token_positions_per_second": 28417.53613224554,
+      "required_pages_per_second_upper_estimate": 0.00220933843875058
     },
     {
       "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": 28673.069882814158,
-      "required_pages_per_second_upper_estimate": 0.3677873537129354
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28417.53613224554,
+      "required_pages_per_second_upper_estimate": 0.36450964113140594
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 0.16419922164885747,
-    "producer_p95_seconds": 0.32828900899767177,
-    "gpu_update_seconds": 0.9642497337395827,
-    "producer_batches_per_second": 6.090162851919695,
-    "gpu_batches_per_second": 1.0370757336087297,
-    "producer_headroom": 5.8724379083942635,
+    "producer_batch_seconds": 0.14678045088334102,
+    "producer_p95_seconds": 0.32799151600193,
+    "gpu_update_seconds": 0.9729203781543769,
+    "producer_batches_per_second": 6.812896363118448,
+    "gpu_batches_per_second": 1.02783333811652,
+    "producer_headroom": 6.62840570593178,
     "sustainable": true,
-    "buffer_seconds": 60.747733225593706,
+    "buffer_seconds": 60.32106344557137,
     "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": 354963.9155388895,
-    "upload_bytes_per_second": 4630086.324175555,
+    "generation_bytes_per_second": 355522.94972227485,
+    "upload_bytes_per_second": 4579609.749604376,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.16699313759817,
+    "estimated_full_shards_per_hour": 76.28694885970292,
     "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": 1.130367224002839,
-      "bytes": 6468621,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1125211710022995,
       "bytes": 2380554,
@@ -562,6 +556,12 @@
       "bytes": 16406792,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0092338640024536,
+      "bytes": 1241016,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -418,3 +418,4 @@
 {"step": 761800, "loss": 0.5890706777572632, "lr": 5e-05, "grad_norm": 0.27169946887613655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 152226.03419675006, "data_wait_s": 1.207699824590236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761900, "loss": 0.5406587719917297, "lr": 5e-05, "grad_norm": 0.2462840110372524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111519760426, "tokens_per_second": 152427.16208617773, "data_wait_s": 1.0782998288050294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 762000, "loss": 0.5618194341659546, "lr": 5e-05, "grad_norm": 0.25727874525994243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111534488578, "tokens_per_second": 152381.51411472692, "data_wait_s": 1.201800478156656e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 762100, "loss": 0.5437605977058411, "lr": 5e-05, "grad_norm": 0.2541325931069935, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111549176468, "tokens_per_second": 151530.7859179251, "data_wait_s": 1.4659999578725547e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:725ad9670ddd0ff09b5e1e00d542a8f95471ad55c57f6be6c2b6b40188183920
-size 895989
+oid sha256:90397bae696971417b49fefbb9bc78725e7d7ad69dff9e6c78fbec119d0fd4a3
+size 898107