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": 77573.67413124334,
-      "required_pages_per_second_upper_estimate": 0.00536672677212197
+      "required_token_positions_per_second": 77122.03542709992,
+      "required_pages_per_second_upper_estimate": 0.005335481358623156
     },
     {
       "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": 9696.709266405418,
-      "required_pages_per_second_upper_estimate": 0.004059809679528525
+      "required_token_positions_per_second": 9640.25442838749,
+      "required_pages_per_second_upper_estimate": 0.004036173217761494
     },
     {
       "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": 19393.418532810836,
-      "required_pages_per_second_upper_estimate": 0.004039137641545015
+      "required_token_positions_per_second": 19280.50885677498,
+      "required_pages_per_second_upper_estimate": 0.00401562153365514
     },
     {
       "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": 29090.127799216254,
-      "required_pages_per_second_upper_estimate": 0.0017439087435097537
+      "required_token_positions_per_second": 28920.76328516247,
+      "required_pages_per_second_upper_estimate": 0.0017337555994968686
     },
     {
       "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": 29090.127799216254,
-      "required_pages_per_second_upper_estimate": 0.0022616294824394654
+      "required_token_positions_per_second": 28920.76328516247,
+      "required_pages_per_second_upper_estimate": 0.0022484621364275517
     },
     {
       "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": 29090.127799216254,
-      "required_pages_per_second_upper_estimate": 0.3731369248626397
+      "required_token_positions_per_second": 28920.76328516247,
+      "required_pages_per_second_upper_estimate": 0.37096449872580484
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08961336391422492,
-    "producer_p95_seconds": 0.33796395199897233,
-    "gpu_update_seconds": 0.9504255254851405,
-    "producer_batches_per_second": 11.159049904176886,
-    "gpu_batches_per_second": 1.0521602936637824,
-    "producer_headroom": 10.605845869092223,
+    "producer_batch_seconds": 0.45504759182023236,
+    "producer_p95_seconds": 0.33293257799959974,
+    "gpu_update_seconds": 0.9559913660433904,
+    "producer_batches_per_second": 2.1975723374337788,
+    "gpu_batches_per_second": 1.046034551691351,
+    "producer_headroom": 2.1008601808424845,
     "sustainable": true,
-    "buffer_seconds": 60.82723363104899,
+    "buffer_seconds": 60.227456060733594,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352347.8379036894,
-    "upload_bytes_per_second": 5561965.2425649725,
+    "generation_bytes_per_second": 338215.1716541958,
+    "upload_bytes_per_second": 5552514.764983085,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.60564377625477,
+    "estimated_full_shards_per_hour": 72.57310259074598,
     "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": 1.1900551170001563,
-      "bytes": 2519802,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0123512749996735,
       "bytes": 5925157,
@@ -562,6 +556,12 @@
       "bytes": 5596441,
       "pages": 20,
       "shards": 2
+    },
+    {
+      "seconds": 1.0342092059945571,
+      "bytes": 930671,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -350,3 +350,4 @@
 {"step": 520000, "loss": 0.6007126569747925, "lr": 0.0003, "grad_norm": 0.15556849611963117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76110796379, "tokens_per_second": 153679.85633253094, "data_wait_s": 1.1628995707724243e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520100, "loss": 0.5963959693908691, "lr": 0.0003, "grad_norm": 0.15683531833211747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76125428033, "tokens_per_second": 133742.67439068158, "data_wait_s": 1.5700003132224083e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520200, "loss": 0.592900276184082, "lr": 0.0003, "grad_norm": 0.1695769086647062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76139940545, "tokens_per_second": 143883.7605864952, "data_wait_s": 2.0942999981343746e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 520300, "loss": 0.5820107460021973, "lr": 0.0003, "grad_norm": 0.15467962670824772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76154388078, "tokens_per_second": 150642.8708769923, "data_wait_s": 1.507999695604667e-05, "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:fbc779649e0da13e52c2f4d74fb3eff1427b8af36696aa03c6d02c67b4b0333e
-size 751182
+oid sha256:84127a554610557a3ef923bc6fbc66be394ea83dd5cabbe9f9d1f8aef9e0d69d
+size 753300