Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75123.9163420815,
-      "required_pages_per_second_upper_estimate": 0.0065115506687063755
+      "required_token_positions_per_second": 73749.15260894394,
+      "required_pages_per_second_upper_estimate": 0.006392389632624838
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9390.489542760188,
-      "required_pages_per_second_upper_estimate": 0.0037460066789373656
+      "required_token_positions_per_second": 9218.644076117993,
+      "required_pages_per_second_upper_estimate": 0.003677454952975105
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18780.979085520376,
-      "required_pages_per_second_upper_estimate": 0.00434464897749237
+      "required_token_positions_per_second": 18437.288152235986,
+      "required_pages_per_second_upper_estimate": 0.004265142128830861
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28171.468628280563,
-      "required_pages_per_second_upper_estimate": 0.0017154946351893097
+      "required_token_positions_per_second": 27655.932228353977,
+      "required_pages_per_second_upper_estimate": 0.0016841011732442322
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28171.468628280563,
-      "required_pages_per_second_upper_estimate": 0.0021687176386213742
+      "required_token_positions_per_second": 27655.932228353977,
+      "required_pages_per_second_upper_estimate": 0.0021290302194589324
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28171.468628280563,
-      "required_pages_per_second_upper_estimate": 0.34979535652284743
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27655.932228353977,
+      "required_pages_per_second_upper_estimate": 0.3433941198250987
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.24495871825774884,
-    "producer_p95_seconds": 0.37026148100267164,
-    "gpu_update_seconds": 0.9814184828207689,
-    "producer_batches_per_second": 4.082320511441387,
-    "gpu_batches_per_second": 1.0189333271224164,
-    "producer_headroom": 4.006464802726912,
+    "producer_batch_seconds": 0.11472205159407167,
+    "producer_p95_seconds": 0.3844791110022925,
+    "gpu_update_seconds": 0.9997131816679156,
+    "producer_batches_per_second": 8.716719986305367,
+    "gpu_batches_per_second": 1.0002869006204418,
+    "producer_headroom": 8.714219871217647,
     "sustainable": true,
-    "buffer_seconds": 60.847945934887676,
+    "buffer_seconds": 60.982504081742846,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 593617.4727335217,
-    "upload_bytes_per_second": 4664112.175691686,
+    "generation_bytes_per_second": 592970.2160377875,
+    "upload_bytes_per_second": 4622615.155007548,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.68824547054405,
+    "estimated_full_shards_per_hour": 63.618802360774126,
     "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": 1.505151831999683,
-      "bytes": 12552564,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 13.275076857000386,
       "bytes": 1045844,
@@ -562,6 +556,12 @@
       "bytes": 5936931,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0263739810034167,
+      "bytes": 6580240,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 73200, "loss": 0.6195316538214684, "lr": 0.0003, "grad_norm": 0.15515854729561093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10712902577, "tokens_per_second": 149556.26445771017, "data_wait_s": 7.280999852810055e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 73300, "loss": 0.6037278231233358, "lr": 0.0003, "grad_norm": 0.13094053326599112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10727361084, "tokens_per_second": 146893.6334221823, "data_wait_s": 1.1030999303329736e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 73400, "loss": 0.6100993007421494, "lr": 0.0003, "grad_norm": 0.12337211974140161, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10741768291, "tokens_per_second": 146735.5201714556, "data_wait_s": 1.044199598254636e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 73500, "loss": 0.6147768199443817, "lr": 0.0003, "grad_norm": 0.12840102069003204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10756446645, "tokens_per_second": 150517.09152512488, "data_wait_s": 1.0752999514807016e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86f53cdac3e6fe84fc14275093f21404bc4579e88c31bc67f0ea8a9d095b4dba
-size 712275
+oid sha256:8dabb9e3cc02b2c31342f45f733ce6fca82108de4dfa8b22427e724d9795d26d
+size 714393