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": 77139.20420071471,
-      "required_pages_per_second_upper_estimate": 0.00533666913421887
+      "required_token_positions_per_second": 75925.2013156566,
+      "required_pages_per_second_upper_estimate": 0.005252681597755767
     },
     {
       "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": 9642.400525089339,
-      "required_pages_per_second_upper_estimate": 0.004037071743634963
+      "required_token_positions_per_second": 9490.650164457074,
+      "required_pages_per_second_upper_estimate": 0.003973537036546113
     },
     {
       "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": 19284.801050178678,
-      "required_pages_per_second_upper_estimate": 0.004016515484348378
+      "required_token_positions_per_second": 18981.30032891415,
+      "required_pages_per_second_upper_estimate": 0.0039533042879612325
     },
     {
       "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": 28927.201575268016,
-      "required_pages_per_second_upper_estimate": 0.0017341415651579964
+      "required_token_positions_per_second": 28471.950493371223,
+      "required_pages_per_second_upper_estimate": 0.0017068499579264364
     },
     {
       "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": 28927.201575268016,
-      "required_pages_per_second_upper_estimate": 0.0022489626851642124
+      "required_token_positions_per_second": 28471.950493371223,
+      "required_pages_per_second_upper_estimate": 0.002213568915984622
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28927.201575268016,
-      "required_pages_per_second_upper_estimate": 0.37104708219838145
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28471.950493371223,
+      "required_pages_per_second_upper_estimate": 0.36520761013033726
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09885321918750378,
-    "producer_p95_seconds": 0.3680573860037839,
-    "gpu_update_seconds": 0.955778592272759,
-    "producer_batches_per_second": 10.116008443824274,
-    "gpu_batches_per_second": 1.046267418086951,
-    "producer_headroom": 9.668664309857707,
+    "producer_batch_seconds": 0.4147485447969075,
+    "producer_p95_seconds": 0.48474693699972704,
+    "gpu_update_seconds": 0.9710609747806687,
+    "producer_batches_per_second": 2.411099478334942,
+    "gpu_batches_per_second": 1.0298014501364012,
+    "producer_headroom": 2.3413246097250906,
     "sustainable": true,
-    "buffer_seconds": 60.214051313183816,
+    "buffer_seconds": 60.205780436401454,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338392.3849349778,
+    "generation_bytes_per_second": 318752.49178369227,
     "upload_bytes_per_second": 5112154.702180811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.61112843548776,
+    "estimated_full_shards_per_hour": 68.39686455853534,
     "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": [

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

@@ -398,3 +398,4 @@
 {"step": 436900, "loss": 0.6994632482528687, "lr": 0.0003, "grad_norm": 0.16904049871297688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63949527615, "tokens_per_second": 151376.60636436543, "data_wait_s": 1.4575998648069799e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 437000, "loss": 0.5981572866439819, "lr": 0.0003, "grad_norm": 0.14569365186351016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63964172030, "tokens_per_second": 152071.62493948467, "data_wait_s": 1.4822995581198484e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 437100, "loss": 0.6086249351501465, "lr": 0.0003, "grad_norm": 0.15921939448683664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63978818692, "tokens_per_second": 150419.73014183622, "data_wait_s": 1.437800528947264e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 437200, "loss": 0.6126549243927002, "lr": 0.0003, "grad_norm": 0.16073425316825568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63993467293, "tokens_per_second": 150883.4941976456, "data_wait_s": 1.3878001482225955e-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:fdc524391b19cf3b76c1dc830aec6bd6a50e257861c57264d04114281f35eb2c
-size 853629
+oid sha256:e55b1e930f9a9a7ba9be679bd6f842b649c5d68c051db90951c0ef4cb7c6b1cf
+size 855747