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": 77499.05732823367,
-      "required_pages_per_second_upper_estimate": 0.005361564608554924
+      "required_token_positions_per_second": 76723.27863287073,
+      "required_pages_per_second_upper_estimate": 0.005307894438355144
     },
     {
       "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": 9687.382166029209,
-      "required_pages_per_second_upper_estimate": 0.00405590461737301
+      "required_token_positions_per_second": 9590.409829108841,
+      "required_pages_per_second_upper_estimate": 0.004015304324917119
     },
     {
       "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": 19374.764332058417,
-      "required_pages_per_second_upper_estimate": 0.004035252463472642
+      "required_token_positions_per_second": 19180.819658217682,
+      "required_pages_per_second_upper_estimate": 0.003994858902576612
     },
     {
       "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": 29062.14649808763,
-      "required_pages_per_second_upper_estimate": 0.0017422313072320656
+      "required_token_positions_per_second": 28771.22948732652,
+      "required_pages_per_second_upper_estimate": 0.0017247912766414923
     },
     {
       "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": 29062.14649808763,
-      "required_pages_per_second_upper_estimate": 0.002259454059353451
+      "required_token_positions_per_second": 28771.22948732652,
+      "required_pages_per_second_upper_estimate": 0.002236836541375472
     },
     {
       "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": 29062.14649808763,
-      "required_pages_per_second_upper_estimate": 0.3727780107757421
+      "required_token_positions_per_second": 28771.22948732652,
+      "required_pages_per_second_upper_estimate": 0.36904643972404816
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.3458453014217895,
-    "producer_p95_seconds": 0.190810670999781,
-    "gpu_update_seconds": 0.9513406038958381,
-    "producer_batches_per_second": 2.89146620147489,
-    "gpu_batches_per_second": 1.0511482385014332,
-    "producer_headroom": 2.750769202255527,
+    "producer_batch_seconds": 0.17817234434373574,
+    "producer_p95_seconds": 0.19116019299872278,
+    "gpu_update_seconds": 0.9609599760823119,
+    "producer_batches_per_second": 5.612543314077791,
+    "gpu_batches_per_second": 1.040626066526567,
+    "producer_headroom": 5.393429488857134,
     "sustainable": true,
-    "buffer_seconds": 60.88579864933364,
+    "buffer_seconds": 60.540478493185645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429593.0359157588,
-    "upload_bytes_per_second": 4969980.407006703,
+    "generation_bytes_per_second": 336504.00451576,
+    "upload_bytes_per_second": 4953863.36733734,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.18066509346555,
+    "estimated_full_shards_per_hour": 72.20592595676995,
     "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": 1.183017940999889,
-      "bytes": 7545162,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9269013600001017,
       "bytes": 1114116,
@@ -562,6 +556,12 @@
       "bytes": 6459806,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.1259296549997089,
+      "bytes": 5894764,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -95,3 +95,4 @@
 {"step": 406600, "loss": 0.6084566116333008, "lr": 0.0003, "grad_norm": 0.16289168414628516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59515798258, "tokens_per_second": 151487.60826619124, "data_wait_s": 9.319001037511043e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 406700, "loss": 0.6016278862953186, "lr": 0.0003, "grad_norm": 0.16093596745431255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59530322116, "tokens_per_second": 150622.5098614456, "data_wait_s": 1.6726000467315316e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406800, "loss": 0.546454906463623, "lr": 0.0003, "grad_norm": 0.18739730907034083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59544924867, "tokens_per_second": 110664.96534534721, "data_wait_s": 1.964900002349168e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 406900, "loss": 0.5444069504737854, "lr": 0.0003, "grad_norm": 0.1747523611226791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59559556766, "tokens_per_second": 152496.75751867174, "data_wait_s": 8.297998647321947e-06, "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 sha256:236d22ba2a85fd5e19b53f6b55082b6a9bc86ae7572f324bed79c56677224fd7
-size 207177
+oid sha256:ded28d6ada21784059337171d2aceb55aa73e0098c722e206fd290bf75fcfa95
+size 209295