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": 76621.42013920454,
-      "required_pages_per_second_upper_estimate": 0.00530084763141905
+      "required_token_positions_per_second": 76946.07038363535,
+      "required_pages_per_second_upper_estimate": 0.005323307688621124
     },
     {
       "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": 9577.677517400567,
-      "required_pages_per_second_upper_estimate": 0.004009973571885767
+      "required_token_positions_per_second": 9618.258797954419,
+      "required_pages_per_second_upper_estimate": 0.004026964106620162
     },
     {
       "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": 19155.355034801134,
-      "required_pages_per_second_upper_estimate": 0.003989555293066224
+      "required_token_positions_per_second": 19236.517595908837,
+      "required_pages_per_second_upper_estimate": 0.0040064593141442865
     },
     {
       "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": 28733.0325522017,
-      "required_pages_per_second_upper_estimate": 0.0017225014287041024
+      "required_token_positions_per_second": 28854.776393863256,
+      "required_pages_per_second_upper_estimate": 0.0017297997861196299
     },
     {
       "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": 28733.0325522017,
-      "required_pages_per_second_upper_estimate": 0.0022338668976800793
+      "required_token_positions_per_second": 28854.776393863256,
+      "required_pages_per_second_upper_estimate": 0.0022433319458746983
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28733.0325522017,
-      "required_pages_per_second_upper_estimate": 0.3685564904529406
+      "required_token_positions_per_second": 28854.776393863256,
+      "required_pages_per_second_upper_estimate": 0.37011808973542226
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09414027088286048,
-    "producer_p95_seconds": 0.2797411920000741,
-    "gpu_update_seconds": 0.9622374509119276,
-    "producer_batches_per_second": 10.622446596147023,
-    "gpu_batches_per_second": 1.0392445222873878,
-    "producer_headroom": 10.221315935124593,
+    "producer_batch_seconds": 0.284317861835774,
+    "producer_p95_seconds": 0.29169036600069376,
+    "gpu_update_seconds": 0.9581775863589811,
+    "producer_batches_per_second": 3.517190209377749,
+    "gpu_batches_per_second": 1.0436478730419292,
+    "producer_headroom": 3.3700928255870113,
     "sustainable": true,
-    "buffer_seconds": 60.62095940745144,
+    "buffer_seconds": 60.36518794061581,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422115.4038126822,
-    "upload_bytes_per_second": 5503704.79314713,
+    "generation_bytes_per_second": 389257.06692502374,
+    "upload_bytes_per_second": 5606499.621596961,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.57613931451176,
+    "estimated_full_shards_per_hour": 83.52550512135538,
     "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": 0.9565514620007889,
-      "bytes": 2665835,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.161401912002475,
       "bytes": 5953132,
@@ -562,6 +556,12 @@
       "bytes": 5723812,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.186567714001285,
+      "bytes": 11920154,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -224,3 +224,4 @@
 {"step": 507400, "loss": 0.5982879996299744, "lr": 0.0003, "grad_norm": 0.1681867520757786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74266776156, "tokens_per_second": 154267.11232209133, "data_wait_s": 1.3694003428099677e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507500, "loss": 0.5969457626342773, "lr": 0.0003, "grad_norm": 0.14158622507568192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74281256347, "tokens_per_second": 150296.1681045733, "data_wait_s": 1.5696998161729425e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507600, "loss": 0.6018542647361755, "lr": 0.0003, "grad_norm": 0.18386225132672107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74295830453, "tokens_per_second": 151760.04939684746, "data_wait_s": 2.4901000870158896e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 507700, "loss": 0.6152762770652771, "lr": 0.0003, "grad_norm": 0.1614442365486507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74310310597, "tokens_per_second": 150534.92371814157, "data_wait_s": 9.554001735523343e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5415fc88693b8c1e1971fca3ad446706747512457f6375129d57ad9946baffab
-size 481965
+oid sha256:4de162b9fc8ba72bfa003e53316ac4253bd2f644e29305d5c4ea5fc14a9ed9e0
+size 484083