Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75951.3833665632,
-      "required_pages_per_second_upper_estimate": 0.006583273413192449
+      "required_token_positions_per_second": 75169.10620426339,
+      "required_pages_per_second_upper_estimate": 0.0065154676114276425
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9493.9229208204,
-      "required_pages_per_second_upper_estimate": 0.003787267799912398
+      "required_token_positions_per_second": 9396.138275532923,
+      "required_pages_per_second_upper_estimate": 0.003748260042896491
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18987.8458416408,
-      "required_pages_per_second_upper_estimate": 0.004392503960790268
+      "required_token_positions_per_second": 18792.276551065846,
+      "required_pages_per_second_upper_estimate": 0.004347262447317712
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28481.7687624612,
-      "required_pages_per_second_upper_estimate": 0.0017343902853419265
+      "required_token_positions_per_second": 28188.414826598768,
+      "required_pages_per_second_upper_estimate": 0.0017165265697570572
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28481.7687624612,
-      "required_pages_per_second_upper_estimate": 0.0021926054019163477
+      "required_token_positions_per_second": 28188.414826598768,
+      "required_pages_per_second_upper_estimate": 0.002170022203877989
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28481.7687624612,
-      "required_pages_per_second_upper_estimate": 0.3536482456816271
+      "required_token_positions_per_second": 28188.414826598768,
+      "required_pages_per_second_upper_estimate": 0.35000577159068214
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11785942181256814,
-    "producer_p95_seconds": 0.3124006370017014,
-    "gpu_update_seconds": 0.9707262294903503,
-    "producer_batches_per_second": 8.484684419971957,
-    "gpu_batches_per_second": 1.030156566929297,
-    "producer_headroom": 8.236305715414897,
+    "producer_batch_seconds": 0.0910605247266858,
+    "producer_p95_seconds": 0.2728601320013695,
+    "gpu_update_seconds": 0.9808284775882882,
+    "producer_batches_per_second": 10.981706980071293,
+    "gpu_batches_per_second": 1.0195462538555689,
+    "producer_headroom": 10.771170938584003,
     "sustainable": true,
-    "buffer_seconds": 60.18502622840172,
+    "buffer_seconds": 60.811365610473864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283881.7911544228,
-    "upload_bytes_per_second": 5213977.019993049,
+    "generation_bytes_per_second": 252513.20339717652,
+    "upload_bytes_per_second": 5234639.743072447,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.9144239518596,
+    "estimated_full_shards_per_hour": 54.183455242504806,
     "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": 2.1250439309987996,
-      "bytes": 6028843,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 0.9982244909988367,
       "bytes": 6934101,
@@ -562,6 +556,12 @@
       "bytes": 13456782,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 0.7929018860013457,
+      "bytes": 738445,
+      "pages": 21,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 378400, "loss": 0.5520606003701687, "lr": 0.0003, "grad_norm": 0.1735362186521396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55386921035, "tokens_per_second": 122804.24020786358, "data_wait_s": 1.145099668065086e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 378500, "loss": 0.6116420533508062, "lr": 0.0003, "grad_norm": 0.17059345758723035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55401541945, "tokens_per_second": 148714.45944499297, "data_wait_s": 8.55699909152463e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 378600, "loss": 0.5808735452592373, "lr": 0.0003, "grad_norm": 0.14859657668266757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55416197094, "tokens_per_second": 132678.88405871144, "data_wait_s": 1.4684999769087881e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 378700, "loss": 0.5945687368512154, "lr": 0.0003, "grad_norm": 0.15603520400102683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55430738175, "tokens_per_second": 147506.91357522184, "data_wait_s": 1.0185001883655787e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4caf1853726df9f7be915e40630381cfa9dc4a94497ea558118391585e2f8847
-size 503145
+oid sha256:ea054dfeb00dd00ff01368616601613d2dd7c7a89652c4abe0e6cd246bb13b9c
+size 505263