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": 76617.7284251255,
-      "required_pages_per_second_upper_estimate": 0.005300592230073105
+      "required_token_positions_per_second": 74677.06472575771,
+      "required_pages_per_second_upper_estimate": 0.005166332612390672
     },
     {
       "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": 9577.216053140688,
-      "required_pages_per_second_upper_estimate": 0.004009780366436625
+      "required_token_positions_per_second": 9334.633090719713,
+      "required_pages_per_second_upper_estimate": 0.003908215945779259
     },
     {
       "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": 19154.432106281376,
-      "required_pages_per_second_upper_estimate": 0.00398936307139482
+      "required_token_positions_per_second": 18669.266181439427,
+      "required_pages_per_second_upper_estimate": 0.0038883158039360855
     },
     {
       "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": 28731.648159422064,
-      "required_pages_per_second_upper_estimate": 0.0017224184364708084
+      "required_token_positions_per_second": 28003.89927215914,
+      "required_pages_per_second_upper_estimate": 0.001678791001887606
     },
     {
       "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": 28731.648159422064,
-      "required_pages_per_second_upper_estimate": 0.002233759267230763
+      "required_token_positions_per_second": 28003.89927215914,
+      "required_pages_per_second_upper_estimate": 0.00217717999227513
     },
     {
       "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": 28731.648159422064,
-      "required_pages_per_second_upper_estimate": 0.36853873294880857
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28003.89927215914,
+      "required_pages_per_second_upper_estimate": 0.3592039516188753
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10498551018758917,
-    "producer_p95_seconds": 0.2984306190010102,
-    "gpu_update_seconds": 0.9622838149273835,
-    "producer_batches_per_second": 9.525123973900683,
-    "gpu_batches_per_second": 1.039194450210578,
-    "producer_headroom": 9.165872635261428,
+    "producer_batch_seconds": 0.7373629564220039,
+    "producer_p95_seconds": 0.29455181899902527,
+    "gpu_update_seconds": 0.9872910815490268,
+    "producer_batches_per_second": 1.356184211982145,
+    "gpu_batches_per_second": 1.0128725141839967,
+    "producer_headroom": 1.3389485774275665,
     "sustainable": true,
-    "buffer_seconds": 60.62388034042516,
+    "buffer_seconds": 60.224755974490634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340952.2016724168,
-    "upload_bytes_per_second": 4603433.7947612535,
+    "generation_bytes_per_second": 268762.9608704791,
+    "upload_bytes_per_second": 4548382.36199914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.1604055178583,
+    "estimated_full_shards_per_hour": 57.670274921281624,
     "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": 1.460754736999661,
-      "bytes": 10771768,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 0.9620606520002184,
       "bytes": 2617341,
@@ -562,6 +556,12 @@
       "bytes": 1256562,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.874451825999131,
+      "bytes": 7588108,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -293,3 +293,4 @@
 {"step": 871800, "loss": 0.6195017695426941, "lr": 3e-05, "grad_norm": 0.36921799767350544, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127606975044, "tokens_per_second": 153495.28849010146, "data_wait_s": 1.8030001228908077e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 871900, "loss": 0.5731968879699707, "lr": 3e-05, "grad_norm": 0.34762593455913327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127621718746, "tokens_per_second": 152841.65976966496, "data_wait_s": 1.4327000826597214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872000, "loss": 0.5325416922569275, "lr": 3e-05, "grad_norm": 0.2988365468834813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127636444537, "tokens_per_second": 154200.01939884928, "data_wait_s": 1.575900023453869e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 872100, "loss": 0.5009547472000122, "lr": 3e-05, "grad_norm": 0.30272764242403954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127651203827, "tokens_per_second": 132028.77519436993, "data_wait_s": 1.2270000297576189e-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 sha256:0f54fbb7a8ce5529996863c2b92198fbf205064fa2be865f0491d4d162385223
-size 629673
+oid sha256:4b9db805a92db9c24f7ceba5abfc277fda6c3ad82238ba737ec6fde7df2f829a
+size 631791