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": 76824.80162191723,
-      "required_pages_per_second_upper_estimate": 0.00531491803430317
+      "required_token_positions_per_second": 74988.8764363776,
+      "required_pages_per_second_upper_estimate": 0.0051879044700342035
     },
     {
       "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": 9603.100202739653,
-      "required_pages_per_second_upper_estimate": 0.004020617519351203
+      "required_token_positions_per_second": 9373.6095545472,
+      "required_pages_per_second_upper_estimate": 0.003924534577262709
     },
     {
       "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": 19206.200405479307,
-      "required_pages_per_second_upper_estimate": 0.004000145042895892
+      "required_token_positions_per_second": 18747.2191090944,
+      "required_pages_per_second_upper_estimate": 0.0039045513430096706
     },
     {
       "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": 28809.300608218957,
-      "required_pages_per_second_upper_estimate": 0.0017270735822077581
+      "required_token_positions_per_second": 28120.828663641598,
+      "required_pages_per_second_upper_estimate": 0.0016858007403661363
     },
     {
       "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": 28809.300608218957,
-      "required_pages_per_second_upper_estimate": 0.002239796403567696
+      "required_token_positions_per_second": 28120.828663641598,
+      "required_pages_per_second_upper_estimate": 0.002186270738144853
     },
     {
       "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": 28809.300608218957,
-      "required_pages_per_second_upper_estimate": 0.36953477518527156
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28120.828663641598,
+      "required_pages_per_second_upper_estimate": 0.36070379630381344
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.46613480814050945,
-    "producer_p95_seconds": 0.35763232800672995,
-    "gpu_update_seconds": 0.9596900798109742,
-    "producer_batches_per_second": 2.1453021369272314,
-    "gpu_batches_per_second": 1.0420030601931047,
-    "producer_headroom": 2.0588251790063485,
+    "producer_batch_seconds": 0.12070458972686993,
+    "producer_p95_seconds": 0.3576352240052074,
+    "gpu_update_seconds": 0.983185820400345,
+    "producer_batches_per_second": 8.284689109691667,
+    "gpu_batches_per_second": 1.0171017311791666,
+    "producer_headroom": 8.145388859074005,
     "sustainable": true,
-    "buffer_seconds": 60.46047502809137,
+    "buffer_seconds": 60.95752086482139,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268219.6823755835,
+    "generation_bytes_per_second": 267635.6037675032,
     "upload_bytes_per_second": 3903464.288763667,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.553700003153125,
+    "estimated_full_shards_per_hour": 57.42837033051322,
     "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": [

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

@@ -359,3 +359,4 @@
 {"step": 878400, "loss": 0.5285022258758545, "lr": 3e-05, "grad_norm": 0.3028094354470684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128573480394, "tokens_per_second": 153739.61314169862, "data_wait_s": 1.648499892326072e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 878500, "loss": 0.5892738699913025, "lr": 3e-05, "grad_norm": 0.3085443776722021, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128588161800, "tokens_per_second": 151940.98682745785, "data_wait_s": 9.24799678614363e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 878600, "loss": 0.5238515138626099, "lr": 3e-05, "grad_norm": 0.3148988540399598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128602737008, "tokens_per_second": 152306.6153859522, "data_wait_s": 1.9182996766176075e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 878700, "loss": 0.5303773880004883, "lr": 3e-05, "grad_norm": 0.32359358407629607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128617254746, "tokens_per_second": 150282.72802974845, "data_wait_s": 1.752200478222221e-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:fb453001fb7cc5704543e5cb4a4eaf13b22d7fdce7b6103bcc92475c71cb2ae8
-size 770244
+oid sha256:d331919ec9cb43541cd28a991dbd2ded23cd541041393454c83c0aed2a363b46
+size 772362