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": 76570.33670294234,
-      "required_pages_per_second_upper_estimate": 0.0052973135607685285
+      "required_token_positions_per_second": 75589.26263467403,
+      "required_pages_per_second_upper_estimate": 0.005229440580320261
     },
     {
       "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": 9571.292087867792,
-      "required_pages_per_second_upper_estimate": 0.004007300125883326
+      "required_token_positions_per_second": 9448.657829334254,
+      "required_pages_per_second_upper_estimate": 0.003955955722729861
     },
     {
       "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": 19142.584175735585,
-      "required_pages_per_second_upper_estimate": 0.003986895459913072
+      "required_token_positions_per_second": 18897.315658668507,
+      "required_pages_per_second_upper_estimate": 0.00393581249597376
     },
     {
       "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": 28713.87626360338,
-      "required_pages_per_second_upper_estimate": 0.001721353038452592
+      "required_token_positions_per_second": 28345.97348800276,
+      "required_pages_per_second_upper_estimate": 0.0016992978288103465
     },
     {
       "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": 28713.87626360338,
-      "required_pages_per_second_upper_estimate": 0.002232377580501168
+      "required_token_positions_per_second": 28345.97348800276,
+      "required_pages_per_second_upper_estimate": 0.00220377475793151
     },
     {
       "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": 28713.87626360338,
-      "required_pages_per_second_upper_estimate": 0.3683107741512215
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28345.97348800276,
+      "required_pages_per_second_upper_estimate": 0.3635917123690404
     }
   ],
   "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": 0.14394605829704687,
-    "producer_p95_seconds": 0.30916982399867265,
-    "gpu_update_seconds": 0.9628794017979926,
-    "producer_batches_per_second": 6.947046774538289,
-    "gpu_batches_per_second": 1.0385516588398214,
-    "producer_headroom": 6.689168242530101,
+    "producer_batch_seconds": 0.10143420909375322,
+    "producer_p95_seconds": 0.2936707920016488,
+    "gpu_update_seconds": 0.9753766266556986,
+    "producer_batches_per_second": 9.858606962427476,
+    "gpu_batches_per_second": 1.025244990162137,
+    "producer_headroom": 9.615854802536896,
     "sustainable": true,
-    "buffer_seconds": 60.661402313273534,
+    "buffer_seconds": 60.47335085265331,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268936.15114385146,
-    "upload_bytes_per_second": 5553284.686062772,
+    "generation_bytes_per_second": 257361.01656972963,
+    "upload_bytes_per_second": 5486691.433986545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.7074375222841,
+    "estimated_full_shards_per_hour": 55.22368309801976,
     "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": 1.3665102609993482,
-      "bytes": 11866902,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1544846530014183,
       "bytes": 2585533,
@@ -562,6 +556,12 @@
       "bytes": 11067407,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.5320398900003056,
+      "bytes": 7616229,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -321,3 +321,4 @@
 {"step": 472300, "loss": 0.5671371221542358, "lr": 0.0003, "grad_norm": 0.16899740923244383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69129942936, "tokens_per_second": 153977.98346577233, "data_wait_s": 1.2251002772245556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 472400, "loss": 0.6156185269355774, "lr": 0.0003, "grad_norm": 0.16784349410714736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69144632620, "tokens_per_second": 122424.27870087356, "data_wait_s": 8.983999578049406e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 472500, "loss": 0.5736663341522217, "lr": 0.0003, "grad_norm": 0.16556079692463208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69159366715, "tokens_per_second": 153450.70044017528, "data_wait_s": 1.2171996786491945e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 472600, "loss": 0.5829270482063293, "lr": 0.0003, "grad_norm": 0.15995475005757198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69174063494, "tokens_per_second": 152817.12225836012, "data_wait_s": 1.0965999535983428e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f4273d6fdaab68eebe978c0686fa884a9ae17ff929155e56d2e1983acf69c46
-size 688977
+oid sha256:4344f50afee5c8450f67515637fbefb5cae47183b2497f6c5a8ca7274c9a47f0
+size 691095