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": 76361.37578266983,
-      "required_pages_per_second_upper_estimate": 0.005282857159447941
+      "required_token_positions_per_second": 75743.07913097787,
+      "required_pages_per_second_upper_estimate": 0.005240081962438004
     },
     {
       "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": 9545.171972833728,
-      "required_pages_per_second_upper_estimate": 0.00399636417916859
+      "required_token_positions_per_second": 9467.884891372234,
+      "required_pages_per_second_upper_estimate": 0.003964005692098451
     },
     {
       "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": 19090.343945667457,
-      "required_pages_per_second_upper_estimate": 0.003976015197657392
+      "required_token_positions_per_second": 18935.769782744468,
+      "required_pages_per_second_upper_estimate": 0.003943821475915335
     },
     {
       "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": 28635.515918501187,
-      "required_pages_per_second_upper_estimate": 0.0017166554554130346
+      "required_token_positions_per_second": 28403.6546741167,
+      "required_pages_per_second_upper_estimate": 0.0017027557278438056
     },
     {
       "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": 28635.515918501187,
-      "required_pages_per_second_upper_estimate": 0.002226285408340205
+      "required_token_positions_per_second": 28403.6546741167,
+      "required_pages_per_second_upper_estimate": 0.002208259216439144
     },
     {
       "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": 28635.515918501187,
-      "required_pages_per_second_upper_estimate": 0.3673056517810339
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28403.6546741167,
+      "required_pages_per_second_upper_estimate": 0.3643315846912777
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1115178146875877,
-    "producer_p95_seconds": 0.2265616990007402,
-    "gpu_update_seconds": 0.9655142962567278,
-    "producer_batches_per_second": 8.967177152829406,
-    "gpu_batches_per_second": 1.0357174449689377,
-    "producer_headroom": 8.657937738123492,
+    "producer_batch_seconds": 0.08605613933589495,
+    "producer_p95_seconds": 0.22428638800010958,
+    "gpu_update_seconds": 0.9733958646242343,
+    "producer_batches_per_second": 11.620321428745399,
+    "gpu_batches_per_second": 1.027331259914522,
+    "producer_headroom": 11.311172824345146,
     "sustainable": true,
-    "buffer_seconds": 60.82740066417385,
+    "buffer_seconds": 60.35054360670253,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 502071.9960687514,
-    "upload_bytes_per_second": 3451354.132425653,
+    "generation_bytes_per_second": 551112.5693575254,
+    "upload_bytes_per_second": 3495788.8247184367,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.73296271845727,
+    "estimated_full_shards_per_hour": 118.25592814010928,
     "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": 1.0337117700000817,
-      "bytes": 1192866,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3218913410000823,
       "bytes": 7396091,
@@ -562,6 +556,12 @@
       "bytes": 6647922,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.248109399999521,
+      "bytes": 7422482,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -154,3 +154,4 @@
 {"step": 857900, "loss": 0.48786690831184387, "lr": 3e-05, "grad_norm": 0.30166974147862247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125572644208, "tokens_per_second": 154451.55907358092, "data_wait_s": 1.5167001038207673e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858000, "loss": 0.49318116903305054, "lr": 3e-05, "grad_norm": 0.31133617668927394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125587339788, "tokens_per_second": 153133.82451254097, "data_wait_s": 1.4123999790172093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 858100, "loss": 0.5367608666419983, "lr": 3e-05, "grad_norm": 0.30395223893279827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125602043141, "tokens_per_second": 153264.8200701733, "data_wait_s": 1.3453998690238222e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 858200, "loss": 0.4901749789714813, "lr": 3e-05, "grad_norm": 0.3019906820753487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125616557291, "tokens_per_second": 150899.9502123738, "data_wait_s": 1.3232000128482468e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e718c89b9a00980f72c63c9308724c9b0c0e537c28b157c74c6d908aee80f8c5
-size 332922
+oid sha256:daa2749fd4b558435bfbed5dfa37cac7d5962553a17d3c35501230ce28bf230e
+size 335040