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": 75655.11393090448,
-      "required_pages_per_second_upper_estimate": 0.0065575935293975705
+      "required_token_positions_per_second": 74757.2011996098,
+      "required_pages_per_second_upper_estimate": 0.006479764729587959
     },
     {
       "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": 9456.88924136306,
-      "required_pages_per_second_upper_estimate": 0.0037724945114740146
+      "required_token_positions_per_second": 9344.650149951225,
+      "required_pages_per_second_upper_estimate": 0.0037277206597858725
     },
     {
       "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": 18913.77848272612,
-      "required_pages_per_second_upper_estimate": 0.0043753697808463525
+      "required_token_positions_per_second": 18689.30029990245,
+      "required_pages_per_second_upper_estimate": 0.004323440703931182
     },
     {
       "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": 28370.667724089184,
-      "required_pages_per_second_upper_estimate": 0.00172762481500717
+      "required_token_positions_per_second": 28033.950449853673,
+      "required_pages_per_second_upper_estimate": 0.0017071204996252339
     },
     {
       "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": 28370.667724089184,
-      "required_pages_per_second_upper_estimate": 0.0021840525364351117
+      "required_token_positions_per_second": 28033.950449853673,
+      "required_pages_per_second_upper_estimate": 0.0021581311085713904
     },
     {
       "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": 28370.667724089184,
-      "required_pages_per_second_upper_estimate": 0.35226874261630287
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28033.950449853673,
+      "required_pages_per_second_upper_estimate": 0.34808784099052203
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.12080289999215665,
-    "producer_p95_seconds": 0.37444960099674063,
-    "gpu_update_seconds": 0.9745276448507564,
-    "producer_batches_per_second": 8.277946970353582,
-    "gpu_batches_per_second": 1.026138155529846,
-    "producer_headroom": 8.06708816521813,
+    "producer_batch_seconds": 0.40631728021094204,
+    "producer_p95_seconds": 0.38507903500430984,
+    "gpu_update_seconds": 0.9862327483761502,
+    "producer_batches_per_second": 2.4611308666981726,
+    "gpu_batches_per_second": 1.013959434673527,
+    "producer_headroom": 2.427247858777115,
     "sustainable": true,
-    "buffer_seconds": 60.420713980746896,
+    "buffer_seconds": 60.16019765094516,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422676.08247588953,
-    "upload_bytes_per_second": 5713830.315093775,
+    "generation_bytes_per_second": 429211.6867792721,
+    "upload_bytes_per_second": 5752909.947063382,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.69644790370478,
+    "estimated_full_shards_per_hour": 92.09883644612906,
     "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": 1.048562693002168,
-      "bytes": 883869,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.072776924003847,
       "bytes": 6196482,
@@ -562,6 +556,12 @@
       "bytes": 5847294,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.8758766960017965,
+      "bytes": 2703346,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -411,3 +411,4 @@
 {"step": 396100, "loss": 0.6351889139041305, "lr": 0.0003, "grad_norm": 0.168936021406292, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57977746469, "tokens_per_second": 146864.2931407333, "data_wait_s": 1.3502001820597798e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396200, "loss": 0.6278204917907715, "lr": 0.0003, "grad_norm": 0.15594749022239351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57992267239, "tokens_per_second": 148375.80718384945, "data_wait_s": 1.005100057227537e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396300, "loss": 0.5455999113619328, "lr": 0.0003, "grad_norm": 0.15656467796505655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58006849836, "tokens_per_second": 148805.33496927447, "data_wait_s": 1.1357995390426368e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 396400, "loss": 0.521472217515111, "lr": 0.0003, "grad_norm": 0.17155483494884516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58021582245, "tokens_per_second": 150481.0228060532, "data_wait_s": 1.0001000191550702e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72bec682c63d5efeb36ddeff5d23e1f2941d3f9f946f5ef362e45c3d5f375336
-size 881163
+oid sha256:824aa583d2017c7bc7387b1a23c747aaaae1d5fb21f0470751fd27b09cd6f7a0
+size 883281