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": 75380.56268723792,
-      "required_pages_per_second_upper_estimate": 0.005215002233706036
+      "required_token_positions_per_second": 76608.48295631792,
+      "required_pages_per_second_upper_estimate": 0.005299952607088553
     },
     {
       "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": 9422.57033590474,
-      "required_pages_per_second_upper_estimate": 0.003945033433999733
+      "required_token_positions_per_second": 9576.06036953974,
+      "required_pages_per_second_upper_estimate": 0.0040092965058463265
     },
     {
       "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": 18845.14067180948,
-      "required_pages_per_second_upper_estimate": 0.003924945822157956
+      "required_token_positions_per_second": 19152.12073907948,
+      "required_pages_per_second_upper_estimate": 0.003988881674561517
     },
     {
       "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": 28267.71100771422,
-      "required_pages_per_second_upper_estimate": 0.0016946061126169368
+      "required_token_positions_per_second": 28728.18110861922,
+      "required_pages_per_second_upper_estimate": 0.0017222105920690581
     },
     {
       "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": 28267.71100771422,
-      "required_pages_per_second_upper_estimate": 0.0021976901943293464
+      "required_token_positions_per_second": 28728.18110861922,
+      "required_pages_per_second_upper_estimate": 0.0022334897192807864
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28267.71100771422,
-      "required_pages_per_second_upper_estimate": 0.36258784530360333
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28728.18110861922,
+      "required_pages_per_second_upper_estimate": 0.3684942613437388
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.6952996298595053,
-    "producer_p95_seconds": 0.3487061680025363,
-    "gpu_update_seconds": 0.9780770714846666,
-    "producer_batches_per_second": 1.4382288686131812,
-    "gpu_batches_per_second": 1.022414315961886,
-    "producer_headroom": 1.4066986799378856,
+    "producer_batch_seconds": 0.08617622139058767,
+    "producer_p95_seconds": 0.22807512899817084,
+    "gpu_update_seconds": 0.962399947823528,
+    "producer_batches_per_second": 11.604129118954638,
+    "gpu_batches_per_second": 1.0390690505142948,
+    "producer_headroom": 11.167813258619427,
     "sustainable": true,
-    "buffer_seconds": 60.64077843204933,
+    "buffer_seconds": 60.631196712882264,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334997.191218758,
-    "upload_bytes_per_second": 5552437.450134637,
+    "generation_bytes_per_second": 334302.0562461318,
+    "upload_bytes_per_second": 4757192.28809234,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.88259890005165,
+    "estimated_full_shards_per_hour": 71.73343911684003,
     "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": 0.9846142869992036,
-      "bytes": 7709639,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0955427600001713,
       "bytes": 6199869,
@@ -562,6 +556,12 @@
       "bytes": 283568,
       "pages": 8,
       "shards": 1
+    },
+    {
+      "seconds": 14.44776459899731,
+      "bytes": 11619395,
+      "pages": 28,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 415000, "loss": 0.6023680567741394, "lr": 0.0003, "grad_norm": 0.15095000167774822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60745164755, "tokens_per_second": 151931.7864307439, "data_wait_s": 1.166299989563413e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415100, "loss": 0.5824700593948364, "lr": 0.0003, "grad_norm": 0.15353316396598302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60759930315, "tokens_per_second": 117552.91918434553, "data_wait_s": 1.4602999726776034e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415200, "loss": 0.5866156816482544, "lr": 0.0003, "grad_norm": 0.16299583387780173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60774386259, "tokens_per_second": 149488.56135730998, "data_wait_s": 1.4161996659822762e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 415300, "loss": 0.586100697517395, "lr": 0.0003, "grad_norm": 0.18045619156312825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60788927038, "tokens_per_second": 150968.94529780687, "data_wait_s": 1.1275998986093327e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74b1ac46883a8f5011c84c86daeaab8a3d8f4faf66b353fb4c9cdb4715f9b816
-size 386655
+oid sha256:706f846f2e02d8f7fdcbd21982619e253a4a944bb84b88bc39068ffc8d27bbe3
+size 388773