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": 75024.00069345278,
-      "required_pages_per_second_upper_estimate": 0.0065028902335171335
+      "required_token_positions_per_second": 74065.2577620236,
+      "required_pages_per_second_upper_estimate": 0.006419788826132596
     },
     {
       "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": 9378.000086681597,
-      "required_pages_per_second_upper_estimate": 0.0037410244481736066
+      "required_token_positions_per_second": 9258.15722025295,
+      "required_pages_per_second_upper_estimate": 0.0036932173369446904
     },
     {
       "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": 18756.000173363194,
-      "required_pages_per_second_upper_estimate": 0.004338870545778645
+      "required_token_positions_per_second": 18516.3144405059,
+      "required_pages_per_second_upper_estimate": 0.004283423469806934
     },
     {
       "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": 28134.00026004479,
-      "required_pages_per_second_upper_estimate": 0.001713213008145087
+      "required_token_positions_per_second": 27774.47166075885,
+      "required_pages_per_second_upper_estimate": 0.0016913196027493523
     },
     {
       "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": 28134.00026004479,
-      "required_pages_per_second_upper_estimate": 0.002165833219915503
+      "required_token_positions_per_second": 27774.47166075885,
+      "required_pages_per_second_upper_estimate": 0.00213815571310361
     },
     {
       "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": 28134.00026004479,
-      "required_pages_per_second_upper_estimate": 0.34933012478792097
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27774.47166075885,
+      "required_pages_per_second_upper_estimate": 0.3448659828496076
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.06881863532817079,
-    "producer_p95_seconds": 0.259249513001123,
-    "gpu_update_seconds": 0.9827255187476842,
-    "producer_batches_per_second": 14.530947834570789,
-    "gpu_batches_per_second": 1.0175781344055552,
-    "producer_headroom": 14.279933248624117,
+    "producer_batch_seconds": 0.11866237890620823,
+    "producer_p95_seconds": 0.2666467070012004,
+    "gpu_update_seconds": 0.9954464782515545,
+    "producer_batches_per_second": 8.427270793133253,
+    "gpu_batches_per_second": 1.004574351155919,
+    "producer_headroom": 8.388897032296681,
     "sustainable": true,
-    "buffer_seconds": 60.92898216235642,
+    "buffer_seconds": 60.72223517334483,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308046.5852740936,
-    "upload_bytes_per_second": 5315984.957597301,
+    "generation_bytes_per_second": 348556.03744131804,
+    "upload_bytes_per_second": 5323381.389409212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.09962624232394,
+    "estimated_full_shards_per_hour": 74.79201166562706,
     "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": 0.9960888349996821,
-      "bytes": 2528980,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.481450629999017,
       "bytes": 11596666,
@@ -562,6 +556,12 @@
       "bytes": 11817435,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 0.9047177330030536,
+      "bytes": 2631324,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 175700, "loss": 0.5578190144151449, "lr": 0.0003, "grad_norm": 0.13688014366409545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25714585331, "tokens_per_second": 130103.14748413797, "data_wait_s": 1.0712999937823042e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 175800, "loss": 0.651535626500845, "lr": 0.0003, "grad_norm": 0.1497352585659833, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25729278247, "tokens_per_second": 149915.1489547573, "data_wait_s": 1.2909000361105427e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 175900, "loss": 0.5911736041307449, "lr": 0.0003, "grad_norm": 0.14925685939116468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25743945074, "tokens_per_second": 149222.00557064585, "data_wait_s": 1.1783002264564857e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 176000, "loss": 0.5919118411839008, "lr": 0.0003, "grad_norm": 0.13506766904797246, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25758535359, "tokens_per_second": 148414.89522368903, "data_wait_s": 1.5083998732734472e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4a2b7a6e13b73d321cbede94d5a88a6c882f966112615329516ba9bcd91124e0
-size 445959
+oid sha256:6bcb460c53dcad402f455873cf78907c71de4beca1d09a71bb7a4b3c3b6dd9ac
+size 448077