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": 77193.33208881175,
-      "required_pages_per_second_upper_estimate": 0.005340413827111427
+      "required_token_positions_per_second": 75946.41279716925,
+      "required_pages_per_second_upper_estimate": 0.00525414905726423
     },
     {
       "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": 9649.166511101468,
-      "required_pages_per_second_upper_estimate": 0.004039904520688387
+      "required_token_positions_per_second": 9493.301599646156,
+      "required_pages_per_second_upper_estimate": 0.003974647137091518
     },
     {
       "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": 19298.333022202936,
-      "required_pages_per_second_upper_estimate": 0.0040193338372589305
+      "required_token_positions_per_second": 18986.603199292313,
+      "required_pages_per_second_upper_estimate": 0.003954408736014907
     },
     {
       "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": 28947.499533304406,
-      "required_pages_per_second_upper_estimate": 0.0017353583967490906
+      "required_token_positions_per_second": 28479.904798938467,
+      "required_pages_per_second_upper_estimate": 0.0017073268063996713
     },
     {
       "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": 28947.499533304406,
-      "required_pages_per_second_upper_estimate": 0.00225054076212026
+      "required_token_positions_per_second": 28479.904798938467,
+      "required_pages_per_second_upper_estimate": 0.0022141873282551824
     },
     {
       "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": 28947.499533304406,
-      "required_pages_per_second_upper_estimate": 0.37130744260982296
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28479.904798938467,
+      "required_pages_per_second_upper_estimate": 0.365309639421486
     }
   ],
   "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": 0.0774018298047423,
-    "producer_p95_seconds": 0.18683953499930794,
-    "gpu_update_seconds": 0.9551084012693629,
-    "producer_batches_per_second": 12.919591210216213,
-    "gpu_batches_per_second": 1.047001574555281,
-    "producer_headroom": 12.33961010584332,
+    "producer_batch_seconds": 0.407757547999978,
+    "producer_p95_seconds": 0.2328139370001736,
+    "gpu_update_seconds": 0.9707897619457817,
+    "producer_batches_per_second": 2.4524377412629867,
+    "gpu_batches_per_second": 1.0300891492671609,
+    "producer_headroom": 2.3808014510275455,
     "sustainable": true,
-    "buffer_seconds": 60.17182927996986,
+    "buffer_seconds": 60.18896524063847,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348880.6074232613,
-    "upload_bytes_per_second": 5539058.967591378,
+    "generation_bytes_per_second": 314547.69782211434,
+    "upload_bytes_per_second": 5594784.931575968,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.86165682814959,
+    "estimated_full_shards_per_hour": 67.49461365697454,
     "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": 0.7817219419998764,
-      "bytes": 1157795,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.0034240229997522,
       "bytes": 6082079,
@@ -562,6 +556,12 @@
       "bytes": 5966085,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.0020495059998211,
+      "bytes": 6673858,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -94,3 +94,4 @@
 {"step": 613800, "loss": 0.5992052555084229, "lr": 0.0001, "grad_norm": 0.14791244464484515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89842313948, "tokens_per_second": 153420.90148336272, "data_wait_s": 1.2147000234108418e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 613900, "loss": 0.5253154635429382, "lr": 0.0001, "grad_norm": 0.15046249175064386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89857060438, "tokens_per_second": 153234.40435261215, "data_wait_s": 1.4196999472915195e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 614000, "loss": 0.5375323295593262, "lr": 0.0001, "grad_norm": 0.16010801183743353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89871650856, "tokens_per_second": 151145.166963365, "data_wait_s": 1.4307999663287774e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 614100, "loss": 0.5311686396598816, "lr": 0.0001, "grad_norm": 0.15072206927000883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89886189019, "tokens_per_second": 150754.44136008003, "data_wait_s": 5.9630001487676054e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c662011db519e74015735d96fa38eec0b552639e3da59bb2e1615362a144da2e
-size 205059
+oid sha256:c5d0da1df028121d6a6a4e9537997401b0ea7ae3e976c4d1e232f4508e469151
+size 207177