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": 76084.80933808602,
-      "required_pages_per_second_upper_estimate": 0.0052637236510890695
+      "required_token_positions_per_second": 76238.26196636575,
+      "required_pages_per_second_upper_estimate": 0.005274339859972614
     },
     {
       "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": 9510.601167260753,
-      "required_pages_per_second_upper_estimate": 0.003981890104795698
+      "required_token_positions_per_second": 9529.78274579572,
+      "required_pages_per_second_upper_estimate": 0.00398992103117138
     },
     {
       "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": 19021.202334521506,
-      "required_pages_per_second_upper_estimate": 0.003961614823442593
+      "required_token_positions_per_second": 19059.56549159144,
+      "required_pages_per_second_upper_estimate": 0.003969604857355775
     },
     {
       "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": 28531.803501782262,
-      "required_pages_per_second_upper_estimate": 0.001710438054390426
+      "required_token_positions_per_second": 28589.348237387156,
+      "required_pages_per_second_upper_estimate": 0.0017138877734242158
     },
     {
       "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": 28531.803501782262,
-      "required_pages_per_second_upper_estimate": 0.002218222224123019
+      "required_token_positions_per_second": 28589.348237387156,
+      "required_pages_per_second_upper_estimate": 0.002222696074203757
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28531.803501782262,
-      "required_pages_per_second_upper_estimate": 0.3659753402570806
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28589.348237387156,
+      "required_pages_per_second_upper_estimate": 0.366713462338697
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12887564699209975,
-    "producer_p95_seconds": 0.28746414400302456,
-    "gpu_update_seconds": 0.9690239174075675,
-    "producer_batches_per_second": 7.759417883359308,
-    "gpu_batches_per_second": 1.0319662724892311,
-    "producer_headroom": 7.519061514135172,
+    "producer_batch_seconds": 0.12863128952349712,
+    "producer_p95_seconds": 0.29422660500131315,
+    "gpu_update_seconds": 0.9670734628306032,
+    "producer_batches_per_second": 7.774158244890561,
+    "gpu_batches_per_second": 1.0340476069656812,
+    "producer_headroom": 7.5181821344794,
     "sustainable": true,
-    "buffer_seconds": 60.079482879269186,
+    "buffer_seconds": 60.925628158328,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353807.39509910916,
-    "upload_bytes_per_second": 4854714.749722261,
+    "generation_bytes_per_second": 380971.5812132714,
+    "upload_bytes_per_second": 4857321.401894118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.91883077363926,
+    "estimated_full_shards_per_hour": 81.74763276384932,
     "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": 1.1927087240001129,
-      "bytes": 6721845,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.142808556996897,
       "bytes": 7320475,
@@ -562,6 +556,12 @@
       "bytes": 10433001,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.281639933000406,
+      "bytes": 7384470,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -314,3 +314,4 @@
 {"step": 791400, "loss": 0.5172816514968872, "lr": 3e-05, "grad_norm": 0.29014122444539825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115838278791, "tokens_per_second": 152309.81981993414, "data_wait_s": 1.1657000868581235e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791500, "loss": 0.5506086349487305, "lr": 3e-05, "grad_norm": 0.28517336271728827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115853113635, "tokens_per_second": 153999.3772493908, "data_wait_s": 1.79190028575249e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 791600, "loss": 0.5225836038589478, "lr": 3e-05, "grad_norm": 0.28814563517960295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115867706785, "tokens_per_second": 151674.07777134108, "data_wait_s": 1.2565997167257592e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 791700, "loss": 0.5417950749397278, "lr": 3e-05, "grad_norm": 0.2542495182302116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115882165321, "tokens_per_second": 149714.58733290606, "data_wait_s": 1.6931000573094934e-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:e3ee396def5511ff172899f91790e1501e4ad076368da695c4100dd900b63069
-size 674151
+oid sha256:3ce4461185589c62d5d5f73c81f93ef85a4991697ac75dfa26c4e2f256cae56e
+size 676269