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": 76121.9822666397,
-      "required_pages_per_second_upper_estimate": 0.0052662953605394905
+      "required_token_positions_per_second": 76915.09341352113,
+      "required_pages_per_second_upper_estimate": 0.005321164629952148
     },
     {
       "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": 9515.247783329962,
-      "required_pages_per_second_upper_estimate": 0.003983835545911499
+      "required_token_positions_per_second": 9614.386676690141,
+      "required_pages_per_second_upper_estimate": 0.004025342930305896
     },
     {
       "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": 19030.495566659923,
-      "required_pages_per_second_upper_estimate": 0.0039635503586180135
+      "required_token_positions_per_second": 19228.773353380282,
+      "required_pages_per_second_upper_estimate": 0.004004846392654997
     },
     {
       "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": 28545.743349989883,
-      "required_pages_per_second_upper_estimate": 0.0017112737270055572
+      "required_token_positions_per_second": 28843.160030070423,
+      "required_pages_per_second_upper_estimate": 0.00172910340284741
     },
     {
       "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": 28545.743349989883,
-      "required_pages_per_second_upper_estimate": 0.0022193059860062493
+      "required_token_positions_per_second": 28843.160030070423,
+      "required_pages_per_second_upper_estimate": 0.00224242882468479
     },
     {
       "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": 28545.743349989883,
-      "required_pages_per_second_upper_estimate": 0.3661541456624451
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28843.160030070423,
+      "required_pages_per_second_upper_estimate": 0.3699690874933675
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10293882089820272,
-    "producer_p95_seconds": 0.25073032299769693,
-    "gpu_update_seconds": 0.9685507103814761,
-    "producer_batches_per_second": 9.71450800848895,
-    "gpu_batches_per_second": 1.0324704626009074,
-    "producer_headroom": 9.408993632628512,
+    "producer_batch_seconds": 0.09313948602346045,
+    "producer_p95_seconds": 0.2501452309988963,
+    "gpu_update_seconds": 0.9585634851096617,
+    "producer_batches_per_second": 10.736584908232315,
+    "gpu_batches_per_second": 1.043227720995024,
+    "producer_headroom": 10.291698247810965,
     "sustainable": true,
-    "buffer_seconds": 60.05014404365152,
+    "buffer_seconds": 60.389499561908686,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324489.1041991831,
-    "upload_bytes_per_second": 5004240.969786775,
+    "generation_bytes_per_second": 328860.8693089413,
+    "upload_bytes_per_second": 4940931.032631615,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.62780804139729,
+    "estimated_full_shards_per_hour": 70.56588706446819,
     "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": 0.9962533819998498,
-      "bytes": 8524201,
-      "pages": 41,
-      "shards": 3
-    },
     {
       "seconds": 1.0446712169996317,
       "bytes": 5880079,
@@ -562,6 +556,12 @@
       "bytes": 10759915,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8831225600006292,
+      "bytes": 2598975,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 420500, "loss": 0.5629096031188965, "lr": 0.0003, "grad_norm": 0.15777687645875652, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61550064028, "tokens_per_second": 152217.03841189836, "data_wait_s": 1.0747997293947265e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420600, "loss": 0.5729127526283264, "lr": 0.0003, "grad_norm": 0.21074862334746303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61564668174, "tokens_per_second": 150827.35342122737, "data_wait_s": 7.499998901039362e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 420700, "loss": 0.6293513774871826, "lr": 0.0003, "grad_norm": 0.17249792678666095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61579348754, "tokens_per_second": 152410.75161214618, "data_wait_s": 1.5664998500142246e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 420800, "loss": 0.5988439917564392, "lr": 0.0003, "grad_norm": 0.17409002378700117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61593903469, "tokens_per_second": 150870.86577655323, "data_wait_s": 1.058699854183942e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd0eb617ee29367f204d834a5ba9addf5e08c57929513a5ef61861607db6183e
-size 503928
+oid sha256:8b4688e18beda04d81a5e21b7ab9c97f2db3f6d4bd6e6f4c6521473366371822
+size 506046