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": 76175.52680510457,
-      "required_pages_per_second_upper_estimate": 0.005269999695950937
+      "required_token_positions_per_second": 76403.13267647855,
+      "required_pages_per_second_upper_estimate": 0.00528574599825097
     },
     {
       "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": 9521.940850638071,
-      "required_pages_per_second_upper_estimate": 0.003986637793426265
+      "required_token_positions_per_second": 9550.391584559819,
+      "required_pages_per_second_upper_estimate": 0.003998549521600417
     },
     {
       "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": 19043.881701276143,
-      "required_pages_per_second_upper_estimate": 0.003966338337442463
+      "required_token_positions_per_second": 19100.783169119637,
+      "required_pages_per_second_upper_estimate": 0.003978189412601646
     },
     {
       "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": 28565.822551914214,
-      "required_pages_per_second_upper_estimate": 0.0017124774445017546
+      "required_token_positions_per_second": 28651.174753679454,
+      "required_pages_per_second_upper_estimate": 0.0017175941786723148
     },
     {
       "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": 28565.822551914214,
-      "required_pages_per_second_upper_estimate": 0.0022208670556367904
+      "required_token_positions_per_second": 28651.174753679454,
+      "required_pages_per_second_upper_estimate": 0.002227502813899378
     },
     {
       "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": 28565.822551914214,
-      "required_pages_per_second_upper_estimate": 0.3664117001053631
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28651.174753679454,
+      "required_pages_per_second_upper_estimate": 0.36750650650555344
     }
   ],
   "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": 0.25456259229673606,
-    "producer_p95_seconds": 0.3294845870004792,
-    "gpu_update_seconds": 0.9678699064153954,
-    "producer_batches_per_second": 3.9283069479208073,
-    "gpu_batches_per_second": 1.0331967068834713,
-    "producer_headroom": 3.802090078055059,
+    "producer_batch_seconds": 0.11838106757045352,
+    "producer_p95_seconds": 0.31961604300158797,
+    "gpu_update_seconds": 0.9649866100673368,
+    "producer_batches_per_second": 8.447296687917248,
+    "gpu_batches_per_second": 1.036283809088522,
+    "producer_headroom": 8.151528195106307,
     "sustainable": true,
-    "buffer_seconds": 60.007934197754516,
+    "buffer_seconds": 60.79415643424222,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 437818.7462051717,
-    "upload_bytes_per_second": 4905450.364779861,
+    "generation_bytes_per_second": 435632.5530285042,
+    "upload_bytes_per_second": 4910529.784657774,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.94571103683818,
+    "estimated_full_shards_per_hour": 93.47660487309784,
     "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": 0.9818654609989608,
-      "bytes": 963058,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.1789819430014177,
       "bytes": 7759537,
@@ -562,6 +556,12 @@
       "bytes": 11976856,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1643081420006638,
+      "bytes": 2305467,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -312,3 +312,4 @@
 {"step": 956200, "loss": 0.4910849332809448, "lr": 3e-05, "grad_norm": 0.3504413896596116, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139961123319, "tokens_per_second": 150464.80419878374, "data_wait_s": 1.5675999748054892e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956300, "loss": 0.5229923129081726, "lr": 3e-05, "grad_norm": 0.343938299650602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139975751263, "tokens_per_second": 152003.12993590246, "data_wait_s": 1.2199001503176987e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956400, "loss": 0.5105358958244324, "lr": 3e-05, "grad_norm": 0.32806969316875106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139990510816, "tokens_per_second": 153783.51315917718, "data_wait_s": 1.1852996976813301e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 956500, "loss": 0.52623450756073, "lr": 3e-05, "grad_norm": 0.3659619853990604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140005236269, "tokens_per_second": 152923.67815141537, "data_wait_s": 1.0313000530004501e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6c94d90002c5f8e9e6ec695dda59d01eaf15911c10f7dd4a6780515be8d0956
-size 669915
+oid sha256:52376f89407f771f3f49ac2e136666d122b83ffe29ef18d962ba564876985d4e
+size 672033