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": 77390.10854074381,
-      "required_pages_per_second_upper_estimate": 0.005354027278640361
+      "required_token_positions_per_second": 76156.3496654344,
+      "required_pages_per_second_upper_estimate": 0.0052686729769314605
     },
     {
       "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": 9673.763567592976,
-      "required_pages_per_second_upper_estimate": 0.004050202794596434
+      "required_token_positions_per_second": 9519.5437081793,
+      "required_pages_per_second_upper_estimate": 0.003985634159936805
     },
     {
       "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": 19347.527135185952,
-      "required_pages_per_second_upper_estimate": 0.0040295796736572916
+      "required_token_positions_per_second": 19039.0874163586,
+      "required_pages_per_second_upper_estimate": 0.003965339814327935
     },
     {
       "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": 29021.29070277893,
-      "required_pages_per_second_upper_estimate": 0.0017397820646864964
+      "required_token_positions_per_second": 28558.631124537904,
+      "required_pages_per_second_upper_estimate": 0.0017120463294112186
     },
     {
       "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": 29021.29070277893,
-      "required_pages_per_second_upper_estimate": 0.0022562777009738516
+      "required_token_positions_per_second": 28558.631124537904,
+      "required_pages_per_second_upper_estimate": 0.0022203079537900282
     },
     {
       "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": 29021.29070277893,
-      "required_pages_per_second_upper_estimate": 0.3722539565010573
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28558.631124537904,
+      "required_pages_per_second_upper_estimate": 0.36631945619653294
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.06694714627352027,
-    "producer_p95_seconds": 0.2514671460012323,
-    "gpu_update_seconds": 0.9526798888153024,
-    "producer_batches_per_second": 14.93715648333067,
-    "gpu_batches_per_second": 1.0496705259975019,
-    "producer_headroom": 14.230328577756236,
+    "producer_batch_seconds": 0.24199974287498094,
+    "producer_p95_seconds": 0.2527878119981324,
+    "gpu_update_seconds": 0.9681136283960236,
+    "producer_batches_per_second": 4.132235795459536,
+    "gpu_batches_per_second": 1.032936600279872,
+    "producer_headroom": 4.00047378933026,
     "sustainable": true,
-    "buffer_seconds": 60.01883299536405,
+    "buffer_seconds": 60.02304496055346,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400506.1695543668,
-    "upload_bytes_per_second": 5526136.612045022,
+    "generation_bytes_per_second": 396946.7626427764,
+    "upload_bytes_per_second": 5462604.031009977,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.93930067990544,
+    "estimated_full_shards_per_hour": 85.17553481543035,
     "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": 1.1459508749994711,
-      "bytes": 7841640,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.0024716899988562,
       "bytes": 5917383,
@@ -562,6 +556,12 @@
       "bytes": 6229682,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.7940784490019723,
+      "bytes": 968621,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -169,3 +169,4 @@
 {"step": 501900, "loss": 0.5525649785995483, "lr": 0.0003, "grad_norm": 0.1397344511084202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73461282059, "tokens_per_second": 153024.28076948592, "data_wait_s": 1.2282998795853928e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502000, "loss": 0.562541663646698, "lr": 0.0003, "grad_norm": 0.16451517640575014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73475920503, "tokens_per_second": 152698.6983736639, "data_wait_s": 1.0954998288070783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502100, "loss": 0.5456511974334717, "lr": 0.0003, "grad_norm": 0.1837708465292662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73490554006, "tokens_per_second": 152031.23970780458, "data_wait_s": 9.715000487631187e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 502200, "loss": 0.626147985458374, "lr": 0.0003, "grad_norm": 0.2118523773437156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73505191055, "tokens_per_second": 136789.732625182, "data_wait_s": 11.220147740998073, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39336eb904ce2e810376354f623b810ff1fdcfa5154083a45f42d298b05ece3f
-size 364692
+oid sha256:535338b74e778f2b167de3f19ff5ddcbd06d053805c7aca637a851f2df3fdd6c
+size 366810