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": 75404.06822750697,
-      "required_pages_per_second_upper_estimate": 0.005216628401522217
+      "required_token_positions_per_second": 77186.27649402899,
+      "required_pages_per_second_upper_estimate": 0.00533992570469312
     },
     {
       "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": 9425.508528438371,
-      "required_pages_per_second_upper_estimate": 0.003946263593857128
+      "required_token_positions_per_second": 9648.284561753624,
+      "required_pages_per_second_upper_estimate": 0.004039535266913642
     },
     {
       "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": 18851.017056876743,
-      "required_pages_per_second_upper_estimate": 0.003926169718196771
+      "required_token_positions_per_second": 19296.569123507248,
+      "required_pages_per_second_upper_estimate": 0.004018966463677756
     },
     {
       "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": 28276.525585315114,
-      "required_pages_per_second_upper_estimate": 0.0016951345330850295
+      "required_token_positions_per_second": 28944.85368526087,
+      "required_pages_per_second_upper_estimate": 0.0017351997821988567
     },
     {
       "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": 28276.525585315114,
-      "required_pages_per_second_upper_estimate": 0.002198375488966586
+      "required_token_positions_per_second": 28944.85368526087,
+      "required_pages_per_second_upper_estimate": 0.0022503350590727307
     },
     {
       "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": 28276.525585315114,
-      "required_pages_per_second_upper_estimate": 0.3627009092407116
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28944.85368526087,
+      "required_pages_per_second_upper_estimate": 0.37127350451201074
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.23353308191400401,
-    "producer_p95_seconds": 0.23206991600090987,
-    "gpu_update_seconds": 0.9777721777232233,
-    "producer_batches_per_second": 4.282048572322781,
-    "gpu_batches_per_second": 1.0227331302558995,
-    "producer_headroom": 4.186867957676665,
+    "producer_batch_seconds": 0.0774071467655233,
+    "producer_p95_seconds": 0.2111015119990043,
+    "gpu_update_seconds": 0.955195707694275,
+    "producer_batches_per_second": 12.918703786216732,
+    "gpu_batches_per_second": 1.0469058769263915,
+    "producer_headroom": 12.339890405568001,
     "sustainable": true,
-    "buffer_seconds": 60.621875018839845,
+    "buffer_seconds": 60.177329584739326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460255.16588832537,
-    "upload_bytes_per_second": 5375136.35211167,
+    "generation_bytes_per_second": 459984.65666842606,
+    "upload_bytes_per_second": 5326530.937953679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.76004440772363,
+    "estimated_full_shards_per_hour": 98.70199942626559,
     "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": 1.2678156700003456,
-      "bytes": 7637773,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.1486175610007194,
       "bytes": 5966085,
@@ -562,6 +556,12 @@
       "bytes": 17697706,
       "pages": 35,
       "shards": 5
+    },
+    {
+      "seconds": 0.7860386789980112,
+      "bytes": 1154184,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -170,3 +170,4 @@
 {"step": 621400, "loss": 0.5217119455337524, "lr": 0.0001, "grad_norm": 0.15606439441137318, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90956389346, "tokens_per_second": 151457.77699158204, "data_wait_s": 1.5958001313265413e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 621500, "loss": 0.5478068590164185, "lr": 0.0001, "grad_norm": 0.16535951936946752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90970908448, "tokens_per_second": 150605.32052202022, "data_wait_s": 1.3943998055765405e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 621600, "loss": 0.5672624707221985, "lr": 0.0001, "grad_norm": 0.15244863120421648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90985525084, "tokens_per_second": 151131.1363444428, "data_wait_s": 1.2983997294213623e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 621700, "loss": 0.49639052152633667, "lr": 0.0001, "grad_norm": 0.17698319896955775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91000274262, "tokens_per_second": 153517.07150469164, "data_wait_s": 1.2032000086037442e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a611052fc7bf93639004f1d781e71ae0cddb9e99bae3d5d59af8d2e4bc2b2ad9
-size 367593
+oid sha256:51212827094cc14cfaa34b131a19e67cdf60af54e752bb166fe93deb93c70e6f
+size 369711