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": 77145.72717431393,
-      "required_pages_per_second_upper_estimate": 0.005337120408667848
+      "required_token_positions_per_second": 77199.2656612594,
+      "required_pages_per_second_upper_estimate": 0.005340824325421141
     },
     {
       "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": 9643.215896789241,
-      "required_pages_per_second_upper_estimate": 0.004037413122738815
+      "required_token_positions_per_second": 9649.908207657425,
+      "required_pages_per_second_upper_estimate": 0.004040215053547981
     },
     {
       "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": 19286.431793578482,
-      "required_pages_per_second_upper_estimate": 0.004016855125192962
+      "required_token_positions_per_second": 19299.81641531485,
+      "required_pages_per_second_upper_estimate": 0.004019642788924436
     },
     {
       "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": 28929.64769036772,
-      "required_pages_per_second_upper_estimate": 0.0017342882060232222
+      "required_token_positions_per_second": 28949.724622972277,
+      "required_pages_per_second_upper_estimate": 0.0017354917874771642
     },
     {
       "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": 28929.64769036772,
-      "required_pages_per_second_upper_estimate": 0.002249152859853891
+      "required_token_positions_per_second": 28949.724622972277,
+      "required_pages_per_second_upper_estimate": 0.0022507137530548018
     },
     {
       "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": 28929.64769036772,
-      "required_pages_per_second_upper_estimate": 0.3710784583364467
+      "required_token_positions_per_second": 28949.724622972277,
+      "required_pages_per_second_upper_estimate": 0.37133598367096726
     }
   ],
   "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": 0.13469518042990103,
-    "producer_p95_seconds": 0.3127284850052092,
-    "gpu_update_seconds": 0.955697777446683,
-    "producer_batches_per_second": 7.4241706110667165,
-    "gpu_batches_per_second": 1.0463558915786937,
-    "producer_headroom": 7.095263352381444,
+    "producer_batch_seconds": 0.4108542318047057,
+    "producer_p95_seconds": 0.30063446500571445,
+    "gpu_update_seconds": 0.9550349911812519,
+    "producer_batches_per_second": 2.4339532675796733,
+    "gpu_batches_per_second": 1.0470820537822727,
+    "producer_headroom": 2.3245105374385324,
     "sustainable": true,
-    "buffer_seconds": 60.208959979141035,
+    "buffer_seconds": 60.16720444441887,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324893.47280671203,
-    "upload_bytes_per_second": 5321755.656743877,
+    "generation_bytes_per_second": 296675.34095897665,
+    "upload_bytes_per_second": 5368689.652992755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.71457613135358,
+    "estimated_full_shards_per_hour": 63.65962192131972,
     "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": 1.2696354079998855,
-      "bytes": 6070118,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9898074049997376,
       "bytes": 2654933,
@@ -562,6 +556,12 @@
       "bytes": 858963,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.6026845509986742,
+      "bytes": 11607409,
+      "pages": 29,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -371,3 +371,4 @@
 {"step": 522100, "loss": 0.5927833318710327, "lr": 0.0003, "grad_norm": 0.15929691437666202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76417899080, "tokens_per_second": 154297.78805688574, "data_wait_s": 1.1994001397397369e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522200, "loss": 0.5561525821685791, "lr": 0.0003, "grad_norm": 0.15404037372103327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76432602826, "tokens_per_second": 153506.2498205698, "data_wait_s": 1.4517005183734e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 522300, "loss": 0.5395720601081848, "lr": 0.0003, "grad_norm": 0.1544161777385268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76447122635, "tokens_per_second": 151519.94599900013, "data_wait_s": 1.3866003428120166e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 522400, "loss": 0.5688287019729614, "lr": 0.0003, "grad_norm": 0.17673751986089603, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76461673859, "tokens_per_second": 146355.84917063563, "data_wait_s": 1.2958997103851289e-05, "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 sha256:69436de78fd038381dcb8ab3b7059d9d136218491b0b602f653ad93e56cc9696
-size 795660
+oid sha256:5484601454aaf04b18e346c28e5ec1c17f7e15a3ed1b6d331799d331152e2f29
+size 797778