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": 74928.37869457439,
-      "required_pages_per_second_upper_estimate": 0.006494601961006096
+      "required_token_positions_per_second": 75125.8121451474,
+      "required_pages_per_second_upper_estimate": 0.006511714992111259
     },
     {
       "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": 9366.047336821799,
-      "required_pages_per_second_upper_estimate": 0.003736256317544997
+      "required_token_positions_per_second": 9390.726518143425,
+      "required_pages_per_second_upper_estimate": 0.0037461012119608364
     },
     {
       "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": 18732.094673643598,
-      "required_pages_per_second_upper_estimate": 0.004333340429140942
+      "required_token_positions_per_second": 18781.45303628685,
+      "required_pages_per_second_upper_estimate": 0.004344758617660945
     },
     {
       "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": 28098.142010465395,
-      "required_pages_per_second_upper_estimate": 0.0017110294288793975
+      "required_token_positions_per_second": 28172.179554430273,
+      "required_pages_per_second_upper_estimate": 0.001715537926861886
     },
     {
       "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": 28098.142010465395,
-      "required_pages_per_second_upper_estimate": 0.0021630727526008933
+      "required_token_positions_per_second": 28172.179554430273,
+      "required_pages_per_second_upper_estimate": 0.0021687723676843485
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28098.142010465395,
-      "required_pages_per_second_upper_estimate": 0.34888488533798623
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28172.179554430273,
+      "required_pages_per_second_upper_estimate": 0.34980418384631007
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08406453481251219,
-    "producer_p95_seconds": 0.17616795799949614,
-    "gpu_update_seconds": 0.983979652095938,
-    "producer_batches_per_second": 11.895622835840157,
-    "gpu_batches_per_second": 1.0162811780405598,
-    "producer_headroom": 11.705050819474494,
+    "producer_batch_seconds": 0.07215428462500029,
+    "producer_p95_seconds": 0.1846224169994457,
+    "gpu_update_seconds": 0.9813937166835981,
+    "producer_batches_per_second": 13.85919083249446,
+    "gpu_batches_per_second": 1.0189590405971598,
+    "producer_headroom": 13.601322801328989,
     "sustainable": true,
-    "buffer_seconds": 60.02275877785222,
+    "buffer_seconds": 60.84641043438308,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367153.7227468702,
-    "upload_bytes_per_second": 5743694.839952308,
+    "generation_bytes_per_second": 389239.6066069598,
+    "upload_bytes_per_second": 5795890.347918879,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.78264200024203,
+    "estimated_full_shards_per_hour": 83.52175854355426,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 9425496,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.236358336000194,
+      "bytes": 7782594,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -41,3 +41,4 @@
 {"step": 199100, "loss": 0.5936685418710113, "lr": 0.0003, "grad_norm": 0.1579531949293927, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29140523105, "tokens_per_second": 147785.8147173087, "data_wait_s": 1.7545000446261838e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199200, "loss": 0.5824197605252266, "lr": 0.0003, "grad_norm": 0.13886453172844473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29154937790, "tokens_per_second": 145347.49475511748, "data_wait_s": 1.88409994734684e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 199300, "loss": 0.6068560369312763, "lr": 0.0003, "grad_norm": 0.15330214921438037, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29169507970, "tokens_per_second": 147532.721754697, "data_wait_s": 1.692100067884894e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 199400, "loss": 0.5781487785279751, "lr": 0.0003, "grad_norm": 0.1592378748404679, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29184147161, "tokens_per_second": 147649.23572227385, "data_wait_s": 1.5172000530583318e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad4041263c4dd8166a0779b297b2bda34ab490268980007b530bc61fd82edb9a
-size 91239
+oid sha256:d4dace91f842c3afbac9dc36ecc7ca638f091c09e4f451291cf9ba55e71fce99
+size 93357