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": 75985.67638442713,
-      "required_pages_per_second_upper_estimate": 0.005256865403334804
+      "required_token_positions_per_second": 76285.96550921143,
+      "required_pages_per_second_upper_estimate": 0.005277640101754147
     },
     {
       "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": 9498.209548053392,
-      "required_pages_per_second_upper_estimate": 0.003976701992600011
+      "required_token_positions_per_second": 9535.745688651428,
+      "required_pages_per_second_upper_estimate": 0.003992417590824659
     },
     {
       "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": 18996.419096106783,
-      "required_pages_per_second_upper_estimate": 0.003956453128458755
+      "required_token_positions_per_second": 19071.491377302857,
+      "required_pages_per_second_upper_estimate": 0.003972088704842707
     },
     {
       "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": 28494.628644160177,
-      "required_pages_per_second_upper_estimate": 0.001708209478438703
+      "required_token_positions_per_second": 28607.237065954287,
+      "required_pages_per_second_upper_estimate": 0.0017149601813821556
     },
     {
       "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": 28494.628644160177,
-      "required_pages_per_second_upper_estimate": 0.00221533204245782
+      "required_token_positions_per_second": 28607.237065954287,
+      "required_pages_per_second_upper_estimate": 0.0022240868519402103
     },
     {
       "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": 28494.628644160177,
-      "required_pages_per_second_upper_estimate": 0.36549850109875676
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28607.237065954287,
+      "required_pages_per_second_upper_estimate": 0.36694292102402853
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 720750,
-    "producer_batch_seconds": 0.05179127841405062,
-    "producer_p95_seconds": 0.18183907999991789,
-    "gpu_update_seconds": 0.9702881320289223,
-    "producer_batches_per_second": 19.30827024591667,
-    "gpu_batches_per_second": 1.030621695752321,
-    "producer_headroom": 18.734585469620107,
+    "producer_batch_seconds": 0.06862940501562775,
+    "producer_p95_seconds": 0.18554538799980946,
+    "gpu_update_seconds": 0.9664687273453653,
+    "producer_batches_per_second": 14.571013689719265,
+    "gpu_batches_per_second": 1.0346946276748512,
+    "producer_headroom": 14.082429056834874,
     "sustainable": true,
-    "buffer_seconds": 60.15786418579318,
+    "buffer_seconds": 60.88752982275801,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357774.0871015351,
-    "upload_bytes_per_second": 4004705.932807293,
+    "generation_bytes_per_second": 357210.0096184099,
+    "upload_bytes_per_second": 3998177.236528712,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.769990537496,
+    "estimated_full_shards_per_hour": 76.6489526406691,
     "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": [
@@ -292,6 +292,12 @@
       "bytes": 7589375,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.573228470999311,
+      "bytes": 6089737,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -62,3 +62,4 @@
 {"step": 806200, "loss": 0.5670785903930664, "lr": 3e-05, "grad_norm": 0.3036575457163, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118004391654, "tokens_per_second": 152700.3048451993, "data_wait_s": 1.4807999832555652e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806300, "loss": 0.5046426653862, "lr": 3e-05, "grad_norm": 0.27817865482446136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118018968740, "tokens_per_second": 150404.6111798633, "data_wait_s": 1.2620000234164763e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806400, "loss": 0.5340363383293152, "lr": 3e-05, "grad_norm": 0.2616473244955081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118033425331, "tokens_per_second": 149681.91922811183, "data_wait_s": 2.4187999770219903e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 806500, "loss": 0.5248042941093445, "lr": 3e-05, "grad_norm": 0.2763862277188099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118047954900, "tokens_per_second": 149281.52852920647, "data_wait_s": 1.666000025579706e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6916c80531301e4433de29bbc2ec90c96df0ba6e541f29fbc7d5b586ea97ed02
-size 136500
+oid sha256:6f93952cb50421e009d762f4413d91fdd25c86e151ea43fce24a8b0f109a6d48
+size 138618