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": 76388.35052627006,
-      "required_pages_per_second_upper_estimate": 0.005284723334800233
+      "required_token_positions_per_second": 77053.7277135633,
+      "required_pages_per_second_upper_estimate": 0.005330755672504961
     },
     {
       "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": 9548.543815783758,
-      "required_pages_per_second_upper_estimate": 0.00399777589939968
+      "required_token_positions_per_second": 9631.715964195413,
+      "required_pages_per_second_upper_estimate": 0.0040325983411076795
     },
     {
       "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": 19097.087631567516,
-      "required_pages_per_second_upper_estimate": 0.003977419729587418
+      "required_token_positions_per_second": 19263.431928390826,
+      "required_pages_per_second_upper_estimate": 0.004012064859821606
     },
     {
       "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": 28645.631447351276,
-      "required_pages_per_second_upper_estimate": 0.0017172618659220791
+      "required_token_positions_per_second": 28895.147892586243,
+      "required_pages_per_second_upper_estimate": 0.0017322199958243636
     },
     {
       "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": 28645.631447351276,
-      "required_pages_per_second_upper_estimate": 0.0022270718462147903
+      "required_token_positions_per_second": 28895.147892586243,
+      "required_pages_per_second_upper_estimate": 0.0022464706523249544
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28645.631447351276,
-      "required_pages_per_second_upper_estimate": 0.3674354029239142
+      "required_token_positions_per_second": 28895.147892586243,
+      "required_pages_per_second_upper_estimate": 0.3706359319735027
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09420027481246507,
-    "producer_p95_seconds": 0.3358083080020151,
-    "gpu_update_seconds": 0.9651733476643776,
-    "producer_batches_per_second": 10.615680283213726,
-    "gpu_batches_per_second": 1.036083313344592,
-    "producer_headroom": 10.24597167668412,
+    "producer_batch_seconds": 0.15130693903114434,
+    "producer_p95_seconds": 0.37083707300189417,
+    "gpu_update_seconds": 0.956838847227142,
+    "producer_batches_per_second": 6.609082216607161,
+    "gpu_batches_per_second": 1.0451080690316203,
+    "producer_headroom": 6.3238266093678,
     "sustainable": true,
-    "buffer_seconds": 60.80592090285579,
+    "buffer_seconds": 60.280847375309946,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354075.3034056676,
-    "upload_bytes_per_second": 4113587.939141124,
+    "generation_bytes_per_second": 369362.01326889056,
+    "upload_bytes_per_second": 4183866.9114862867,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.97631765964051,
+    "estimated_full_shards_per_hour": 79.25648973989523,
     "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": 1.4991452859976562,
-      "bytes": 1121499,
-      "pages": 36,
-      "shards": 1
-    },
     {
       "seconds": 2.2660662760026753,
       "bytes": 11276737,
@@ -562,6 +556,12 @@
       "bytes": 5948663,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.2626624109980185,
+      "bytes": 7374458,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -397,3 +397,4 @@
 {"step": 839700, "loss": 0.5751197934150696, "lr": 3e-05, "grad_norm": 0.30152514482387405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122908383739, "tokens_per_second": 152343.37172487428, "data_wait_s": 8.843999239616096e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 839800, "loss": 0.544115424156189, "lr": 3e-05, "grad_norm": 0.2890185941371288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122923145042, "tokens_per_second": 152966.00068261506, "data_wait_s": 1.6869002138264477e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 839900, "loss": 0.5274035930633545, "lr": 3e-05, "grad_norm": 0.29764063434520743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122937692789, "tokens_per_second": 150850.6700757115, "data_wait_s": 8.830000297166407e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 840000, "loss": 0.549586832523346, "lr": 3e-05, "grad_norm": 0.2927775828313658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 122952232178, "tokens_per_second": 103830.3733807899, "data_wait_s": 1.0750998626463115e-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 sha256:17d7b9a9d078fdf3eb8fd65adb5f36d27655085b4caa59e54145939e6bef486f
-size 850728
+oid sha256:7c2d1512c1bd013ba7d76f52dbb0f0d0290881c2e8858e6637f1ec053509784c
+size 852846