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": 77146.82278846744,
-      "required_pages_per_second_upper_estimate": 0.005337196205797163
+      "required_token_positions_per_second": 77738.47119411839,
+      "required_pages_per_second_upper_estimate": 0.005378127815313534
     },
     {
       "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": 9643.35284855843,
-      "required_pages_per_second_upper_estimate": 0.004037470461584696
+      "required_token_positions_per_second": 9717.308899264799,
+      "required_pages_per_second_upper_estimate": 0.004068434315637497
     },
     {
       "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": 19286.70569711686,
-      "required_pages_per_second_upper_estimate": 0.004016912172076684
+      "required_token_positions_per_second": 19434.617798529598,
+      "required_pages_per_second_upper_estimate": 0.004047718362096532
     },
     {
       "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": 28930.058545675292,
-      "required_pages_per_second_upper_estimate": 0.0017343128361715714
+      "required_token_positions_per_second": 29151.926697794395,
+      "required_pages_per_second_upper_estimate": 0.0017476134931180588
     },
     {
       "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": 28930.058545675292,
-      "required_pages_per_second_upper_estimate": 0.0022491848020468933
+      "required_token_positions_per_second": 29151.926697794395,
+      "required_pages_per_second_upper_estimate": 0.0022664340749792883
     },
     {
       "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": 28930.058545675292,
-      "required_pages_per_second_upper_estimate": 0.37108372834719017
+      "required_token_positions_per_second": 29151.926697794395,
+      "required_pages_per_second_upper_estimate": 0.3739296147791126
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.206808580929561,
+    "producer_batch_seconds": 0.09689756915624059,
     "producer_p95_seconds": 0.21685149899894895,
-    "gpu_update_seconds": 0.9556842049368426,
-    "producer_batches_per_second": 4.835389303022199,
-    "gpu_batches_per_second": 1.0463707517967047,
-    "producer_headroom": 4.6211051816188835,
+    "gpu_update_seconds": 0.9484107272433495,
+    "producer_batches_per_second": 10.320176333707295,
+    "gpu_batches_per_second": 1.054395496882031,
+    "producer_headroom": 9.787765941930939,
     "sustainable": true,
-    "buffer_seconds": 60.20810491102108,
+    "buffer_seconds": 60.69828654357437,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 457544.8213556671,
-    "upload_bytes_per_second": 5872481.822745292,
+    "generation_bytes_per_second": 455557.5798737209,
+    "upload_bytes_per_second": 5849624.474249078,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.17846756460676,
+    "estimated_full_shards_per_hour": 97.75205180319519,
     "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": 1.2498191829999996,
-      "bytes": 5414761,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0708195489996797,
       "bytes": 7566500,
@@ -562,6 +556,12 @@
       "bytes": 10848676,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0698365700009163,
+      "bytes": 2697855,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 495700, "loss": 0.5757331848144531, "lr": 0.0003, "grad_norm": 0.16169313458835852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72553855424, "tokens_per_second": 152365.2462637077, "data_wait_s": 1.3701001080335118e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495800, "loss": 0.5846452713012695, "lr": 0.0003, "grad_norm": 0.15439398962471818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72568502741, "tokens_per_second": 153174.553638269, "data_wait_s": 9.635999958845787e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 495900, "loss": 0.5551076531410217, "lr": 0.0003, "grad_norm": 0.15962606369740412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72583125799, "tokens_per_second": 153637.7892986949, "data_wait_s": 1.7272001059609465e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 496000, "loss": 0.558561384677887, "lr": 0.0003, "grad_norm": 0.14127708065710512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72598066601, "tokens_per_second": 156266.77688317685, "data_wait_s": 1.3274000593810342e-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 sha256:857e6e168f94267a632891bb18fbe22e153591adb5315f158f637e88055cd700
-size 232593
+oid sha256:e82c1c4c280f4d881d51017ec4a7f50a0e8a92c49421f71851fb1871becc6515
+size 234711