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": 76670.34055114374,
-      "required_pages_per_second_upper_estimate": 0.0053042320590279985
+      "required_token_positions_per_second": 77490.03042101624,
+      "required_pages_per_second_upper_estimate": 0.005360940106168313
     },
     {
       "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": 9583.792568892968,
-      "required_pages_per_second_upper_estimate": 0.00401253381624884
+      "required_token_positions_per_second": 9686.25380262703,
+      "required_pages_per_second_upper_estimate": 0.004055432195179425
     },
     {
       "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": 19167.585137785936,
-      "required_pages_per_second_upper_estimate": 0.003992102500988453
+      "required_token_positions_per_second": 19372.50760525406,
+      "required_pages_per_second_upper_estimate": 0.004034782446793182
     },
     {
       "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": 28751.3777066789,
-      "required_pages_per_second_upper_estimate": 0.0017236011927035816
+      "required_token_positions_per_second": 29058.76140788109,
+      "required_pages_per_second_upper_estimate": 0.001742028376242917
     },
     {
       "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": 28751.3777066789,
-      "required_pages_per_second_upper_estimate": 0.0022352931527488745
+      "required_token_positions_per_second": 29058.76140788109,
+      "required_pages_per_second_upper_estimate": 0.0022591908834793313
     },
     {
       "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": 28751.3777066789,
-      "required_pages_per_second_upper_estimate": 0.36879180239708187
+      "required_token_positions_per_second": 29058.76140788109,
+      "required_pages_per_second_upper_estimate": 0.3727345904731993
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10922477356274385,
-    "producer_p95_seconds": 0.33093286600342253,
-    "gpu_update_seconds": 0.9616234839966438,
-    "producer_batches_per_second": 9.155432118387987,
-    "gpu_batches_per_second": 1.0399080478399487,
-    "producer_headroom": 8.804078531179028,
+    "producer_batch_seconds": 0.11979438224994965,
+    "producer_p95_seconds": 0.31162487999972655,
+    "gpu_update_seconds": 0.9514514267115847,
+    "producer_batches_per_second": 8.347636852565516,
+    "gpu_batches_per_second": 1.05102580323644,
+    "producer_headroom": 7.942370993043662,
     "sustainable": true,
-    "buffer_seconds": 60.582279491788555,
+    "buffer_seconds": 60.89289130954142,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348307.9679934998,
-    "upload_bytes_per_second": 5551902.349780062,
+    "generation_bytes_per_second": 391604.23308132886,
+    "upload_bytes_per_second": 5423799.182112922,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.73878173688647,
+    "estimated_full_shards_per_hour": 84.02915233926677,
     "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": 1.198183002001315,
-      "bytes": 11915570,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.2237985769970692,
       "bytes": 7820030,
@@ -562,6 +556,12 @@
       "bytes": 6547762,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2987593140060198,
+      "bytes": 2393188,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 678800, "loss": 0.569308876991272, "lr": 0.0001, "grad_norm": 0.20589001665957765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99358795489, "tokens_per_second": 151683.22996427878, "data_wait_s": 1.487800182076171e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 678900, "loss": 0.6020879149436951, "lr": 0.0001, "grad_norm": 0.2556691093215879, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99373500312, "tokens_per_second": 152638.41731959954, "data_wait_s": 1.2704003893304616e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679000, "loss": 0.5635417103767395, "lr": 0.0001, "grad_norm": 0.1881923509732763, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99388288592, "tokens_per_second": 153790.0230992682, "data_wait_s": 1.0928997653536499e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 679100, "loss": 0.5356208086013794, "lr": 0.0001, "grad_norm": 0.20511929759650416, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99402851670, "tokens_per_second": 151488.33874070636, "data_wait_s": 1.654700463404879e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bbe6a9f02183f57d1776e48b055ecfdeb3fccdc8f52c682a3fab1e72ec2fa878
-size 724983
+oid sha256:11757a6c10e23050f3e9998def4bc8ddbd187d36d60679fd47651aac7bf0034e
+size 727101