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": 77253.80067080278,
-      "required_pages_per_second_upper_estimate": 0.005344597183920007
+      "required_token_positions_per_second": 75802.45286410545,
+      "required_pages_per_second_upper_estimate": 0.005244189575061811
     },
     {
       "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": 9656.725083850348,
-      "required_pages_per_second_upper_estimate": 0.004043069137257395
+      "required_token_positions_per_second": 9475.306608013181,
+      "required_pages_per_second_upper_estimate": 0.003967113009872948
     },
     {
       "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": 19313.450167700696,
-      "required_pages_per_second_upper_estimate": 0.004022482340000178
+      "required_token_positions_per_second": 18950.613216026362,
+      "required_pages_per_second_upper_estimate": 0.003946912971620294
     },
     {
       "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": 28970.175251551045,
-      "required_pages_per_second_upper_estimate": 0.0017367177714341596
+      "required_token_positions_per_second": 28425.91982403954,
+      "required_pages_per_second_upper_estimate": 0.0017040904895847646
     },
     {
       "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": 28970.175251551045,
-      "required_pages_per_second_upper_estimate": 0.0022523036994739924
+      "required_token_positions_per_second": 28425.91982403954,
+      "required_pages_per_second_upper_estimate": 0.0022099902339116004
     },
     {
       "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": 28970.175251551045,
-      "required_pages_per_second_upper_estimate": 0.3715983023762015
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28425.91982403954,
+      "required_pages_per_second_upper_estimate": 0.3646171781280325
     }
   ],
   "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": 0.07902688074227626,
-    "producer_p95_seconds": 0.21254191599837213,
-    "gpu_update_seconds": 0.9543608127990093,
-    "producer_batches_per_second": 12.65392219213632,
-    "gpu_batches_per_second": 1.047821732188623,
-    "producer_headroom": 12.07640746838264,
+    "producer_batch_seconds": 0.27757543203907176,
+    "producer_p95_seconds": 0.2293863609993423,
+    "gpu_update_seconds": 0.9726334335404104,
+    "producer_batches_per_second": 3.6026243124400112,
+    "gpu_batches_per_second": 1.0281365677097636,
+    "producer_headroom": 3.504032854764689,
     "sustainable": true,
-    "buffer_seconds": 60.124731206337586,
+    "buffer_seconds": 60.30327287950545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 451485.569731446,
-    "upload_bytes_per_second": 5303253.270050898,
+    "generation_bytes_per_second": 409934.52064526296,
+    "upload_bytes_per_second": 5341381.462331788,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.87829321820769,
+    "estimated_full_shards_per_hour": 87.96240534323137,
     "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": 1.0513503590000255,
-      "bytes": 1139099,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 0.996597190999637,
       "bytes": 6046442,
@@ -562,6 +556,12 @@
       "bytes": 11976832,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3680221860013262,
+      "bytes": 5930978,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -128,3 +128,4 @@
 {"step": 897800, "loss": 0.5244227051734924, "lr": 3e-05, "grad_norm": 0.31965901024588095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131412877603, "tokens_per_second": 154331.14546529084, "data_wait_s": 1.4981000276748091e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 897900, "loss": 0.5267511606216431, "lr": 3e-05, "grad_norm": 0.323316861854564, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131427537180, "tokens_per_second": 152697.40547576256, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898000, "loss": 0.549811601638794, "lr": 3e-05, "grad_norm": 0.33915004162047574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131441966562, "tokens_per_second": 149895.1024518364, "data_wait_s": 1.588200029800646e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 898100, "loss": 0.5748760104179382, "lr": 3e-05, "grad_norm": 0.3712900950537548, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131456510152, "tokens_per_second": 151927.77316547764, "data_wait_s": 1.4706000001751818e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c90ca596724b875efeec5b53430be7deadb95df8f54171006aa89f056e4d778f
-size 277071
+oid sha256:e561d18c043b7b96229399f8497ef3c7e2050f800c20f2bf9b762b59facac2f8
+size 279189