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": 77357.9470658316,
-      "required_pages_per_second_upper_estimate": 0.005351802273181031
+      "required_token_positions_per_second": 76175.86595422852,
+      "required_pages_per_second_upper_estimate": 0.0052700231590742655
     },
     {
       "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": 9669.74338322895,
-      "required_pages_per_second_upper_estimate": 0.0040485196273542115
+      "required_token_positions_per_second": 9521.983244278565,
+      "required_pages_per_second_upper_estimate": 0.00398665554275826
     },
     {
       "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": 19339.4867664579,
-      "required_pages_per_second_upper_estimate": 0.00402790507689002
+      "required_token_positions_per_second": 19043.96648855713,
+      "required_pages_per_second_upper_estimate": 0.0039663559963971015
     },
     {
       "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": 29009.230149686853,
-      "required_pages_per_second_upper_estimate": 0.0017390590529440246
+      "required_token_positions_per_second": 28565.949732835696,
+      "required_pages_per_second_upper_estimate": 0.0017124850688038035
     },
     {
       "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": 29009.230149686853,
-      "required_pages_per_second_upper_estimate": 0.002255340046019709
+      "required_token_positions_per_second": 28565.949732835696,
+      "required_pages_per_second_upper_estimate": 0.0022208769433940258
     },
     {
       "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": 29009.230149686853,
-      "required_pages_per_second_upper_estimate": 0.3720992566756052
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28565.949732835696,
+      "required_pages_per_second_upper_estimate": 0.3664133314456676
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11779886737500078,
-    "producer_p95_seconds": 0.2626735939993523,
-    "gpu_update_seconds": 0.9530759643512448,
-    "producer_batches_per_second": 8.489045966941271,
-    "gpu_batches_per_second": 1.049234308076058,
-    "producer_headroom": 8.090705671364596,
+    "producer_batch_seconds": 0.23707769977346516,
+    "producer_p95_seconds": 0.27508968799884315,
+    "gpu_update_seconds": 0.967865597278548,
+    "producer_batches_per_second": 4.2180264147810185,
+    "gpu_batches_per_second": 1.0332013068878652,
+    "producer_headroom": 4.0824826552787234,
     "sustainable": true,
-    "buffer_seconds": 60.04378575412842,
+    "buffer_seconds": 60.00766703126998,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 386464.34171536233,
-    "upload_bytes_per_second": 4881637.920145985,
+    "generation_bytes_per_second": 410925.3923611994,
+    "upload_bytes_per_second": 4931459.096727259,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.92625130267766,
+    "estimated_full_shards_per_hour": 88.1750233471583,
     "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": 0.8835518189989671,
-      "bytes": 2669717,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.406538745000944,
       "bytes": 5882576,
@@ -562,6 +556,12 @@
       "bytes": 778500,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.033372367000993,
+      "bytes": 7683491,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -215,3 +215,4 @@
 {"step": 418600, "loss": 0.6041417121887207, "lr": 0.0003, "grad_norm": 0.15793977085570282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61271908979, "tokens_per_second": 152196.6740768357, "data_wait_s": 1.8008999177254736e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418700, "loss": 0.5904332399368286, "lr": 0.0003, "grad_norm": 0.15509538563603148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61286643721, "tokens_per_second": 152560.1140933915, "data_wait_s": 1.0283001756761223e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418800, "loss": 0.627915620803833, "lr": 0.0003, "grad_norm": 0.15717359423047464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61301310935, "tokens_per_second": 151822.85662750484, "data_wait_s": 7.458998879883438e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 418900, "loss": 0.5589568614959717, "lr": 0.0003, "grad_norm": 0.1498780495751591, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61316001869, "tokens_per_second": 152837.25517190874, "data_wait_s": 1.4950001059332862e-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:e51990d7daacf3021241c8abb451c94f6b0a249a389624be6fe9822120a2099b
-size 462903
+oid sha256:16dd24011afe5649b3b48013c28d65b21a27b57a12674238fb5f5db445e502f2
+size 465021