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": 76290.50461106472,
-      "required_pages_per_second_upper_estimate": 0.006612667583170905
+      "required_token_positions_per_second": 75703.02761178675,
+      "required_pages_per_second_upper_estimate": 0.006561746565820332
     },
     {
       "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": 9536.31307638309,
-      "required_pages_per_second_upper_estimate": 0.003804177866755661
+      "required_token_positions_per_second": 9462.878451473343,
+      "required_pages_per_second_upper_estimate": 0.0037748836969336776
     },
     {
       "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": 19072.62615276618,
-      "required_pages_per_second_upper_estimate": 0.004412116393686611
+      "required_token_positions_per_second": 18925.756902946687,
+      "required_pages_per_second_upper_estimate": 0.004378140777551385
     },
     {
       "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": 28608.939229149273,
-      "required_pages_per_second_upper_estimate": 0.0017421342995513564
+      "required_token_positions_per_second": 28388.63535442003,
+      "required_pages_per_second_upper_estimate": 0.0017287189494254475
     },
     {
       "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": 28608.939229149273,
-      "required_pages_per_second_upper_estimate": 0.002202395336472378
+      "required_token_positions_per_second": 28388.63535442003,
+      "required_pages_per_second_upper_estimate": 0.002185435734355564
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28608.939229149273,
-      "required_pages_per_second_upper_estimate": 0.3552272772657198
+      "required_token_positions_per_second": 28388.63535442003,
+      "required_pages_per_second_upper_estimate": 0.3524918404512215
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1345134425780259,
-    "producer_p95_seconds": 0.2968354120021104,
-    "gpu_update_seconds": 0.9664112247765487,
-    "producer_batches_per_second": 7.434201228029234,
-    "gpu_batches_per_second": 1.034756193183929,
-    "producer_headroom": 7.1844955140150555,
+    "producer_batch_seconds": 0.28511397041413034,
+    "producer_p95_seconds": 0.27496239299944136,
+    "gpu_update_seconds": 0.9739108504099083,
+    "producer_batches_per_second": 3.5073693461863407,
+    "gpu_batches_per_second": 1.0267880264185485,
+    "producer_headroom": 3.415865062645983,
     "sustainable": true,
-    "buffer_seconds": 60.88390716092257,
+    "buffer_seconds": 60.38247272541432,
     "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": 419227.1948523787,
-    "upload_bytes_per_second": 5209216.527245422,
+    "generation_bytes_per_second": 396285.5901872087,
+    "upload_bytes_per_second": 5139487.4478438655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.9563969056942,
+    "estimated_full_shards_per_hour": 85.03366259777256,
     "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": 1.1073119590000715,
-      "bytes": 10577372,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.7750392130001273,
       "bytes": 1199165,
@@ -562,6 +556,12 @@
       "bytes": 2639565,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.2386211259981792,
+      "bytes": 5629626,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -160,3 +160,4 @@
 {"step": 171000, "loss": 0.6112257689237595, "lr": 0.0003, "grad_norm": 0.1342801726741275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25026994800, "tokens_per_second": 149090.3955126272, "data_wait_s": 1.2395998055581003e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171100, "loss": 0.6157455686479807, "lr": 0.0003, "grad_norm": 0.14677479063801943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25041665730, "tokens_per_second": 150765.48889829454, "data_wait_s": 1.4601999282604083e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 171200, "loss": 0.5852197669446468, "lr": 0.0003, "grad_norm": 0.13772079506873736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25056408649, "tokens_per_second": 150242.53030800217, "data_wait_s": 9.434999810764566e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 171300, "loss": 0.638681560754776, "lr": 0.0003, "grad_norm": 0.14773664653503873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25071024835, "tokens_per_second": 148637.95263204337, "data_wait_s": 1.2506003258749843e-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 sha256:284c5ad1abeb4e2e4996465993d5a1219f35b53ea247a37331964fa1928cfdc3
-size 345630
+oid sha256:bd7d0fbeee1e9392e00235dfd68c7dcf190ebfbdb4941771acec1606c1a8c5b7
+size 347748