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": 76408.25019020397,
-      "required_pages_per_second_upper_estimate": 0.005286100039724763
+      "required_token_positions_per_second": 77479.7195508929,
+      "required_pages_per_second_upper_estimate": 0.005360226776248664
     },
     {
       "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": 9551.031273775496,
-      "required_pages_per_second_upper_estimate": 0.0039988173461167915
+      "required_token_positions_per_second": 9684.964943861612,
+      "required_pages_per_second_upper_estimate": 0.0040548925769287765
     },
     {
       "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": 19102.06254755099,
-      "required_pages_per_second_upper_estimate": 0.003978455873389419
+      "required_token_positions_per_second": 19369.929887723225,
+      "required_pages_per_second_upper_estimate": 0.004034245576210491
     },
     {
       "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": 28653.093821326485,
-      "required_pages_per_second_upper_estimate": 0.0017177092238475078
+      "required_token_positions_per_second": 29054.894831584836,
+      "required_pages_per_second_upper_estimate": 0.0017417965808978768
     },
     {
       "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": 28653.093821326485,
-      "required_pages_per_second_upper_estimate": 0.0022276520129678487
+      "required_token_positions_per_second": 29054.894831584836,
+      "required_pages_per_second_upper_estimate": 0.002258890274179569
     },
     {
       "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": 28653.093821326485,
-      "required_pages_per_second_upper_estimate": 0.3675311222447953
+      "required_token_positions_per_second": 29054.894831584836,
+      "required_pages_per_second_upper_estimate": 0.37268499418407713
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07567906304684868,
-    "producer_p95_seconds": 0.18989015500119422,
-    "gpu_update_seconds": 0.9649219791903102,
-    "producer_batches_per_second": 13.213694246993462,
-    "gpu_batches_per_second": 1.036353219810709,
-    "producer_headroom": 12.750184005224547,
+    "producer_batch_seconds": 0.08384574994522609,
+    "producer_p95_seconds": 0.21628773500015086,
+    "gpu_update_seconds": 0.9515780442593296,
+    "producer_batches_per_second": 11.926662957314711,
+    "gpu_batches_per_second": 1.050885953109984,
+    "producer_headroom": 11.349150611461726,
     "sustainable": true,
-    "buffer_seconds": 60.790084688989545,
+    "buffer_seconds": 60.90099483259709,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296591.8030502103,
-    "upload_bytes_per_second": 5270270.990680388,
+    "generation_bytes_per_second": 298385.0583468363,
+    "upload_bytes_per_second": 5225438.444681461,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.641696630761444,
+    "estimated_full_shards_per_hour": 64.02648747257058,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 6060902,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.1801391750013863,
+      "bytes": 2749619,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 893600, "loss": 0.5505179762840271, "lr": 3e-05, "grad_norm": 0.3179756803661472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130797547378, "tokens_per_second": 153028.63247185477, "data_wait_s": 7.692000508541241e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 893700, "loss": 0.5575345158576965, "lr": 3e-05, "grad_norm": 0.3109647016489182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130812265732, "tokens_per_second": 153084.95753527686, "data_wait_s": 1.08559997897828e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 893800, "loss": 0.5465090274810791, "lr": 3e-05, "grad_norm": 0.33997115452139703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130826965741, "tokens_per_second": 153941.61458274594, "data_wait_s": 1.633099964237772e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 893900, "loss": 0.5336362719535828, "lr": 3e-05, "grad_norm": 0.31673480207097926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130841703522, "tokens_per_second": 153785.41732130374, "data_wait_s": 1.819099998101592e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:529cb30393eda8f72a6071563a1d03f1c9e7009113be763dad29dfa11d23958c
-size 187332
+oid sha256:cdb5c9151ca5431f4a99b0eeb99aca7a23ab8849c92f9d90a98636545c4dfd8a
+size 189450