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": 77036.26384713945,
-      "required_pages_per_second_upper_estimate": 0.005329547481704005
+      "required_token_positions_per_second": 77006.24757096851,
+      "required_pages_per_second_upper_estimate": 0.0053274708860711435
     },
     {
       "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": 9629.53298089243,
-      "required_pages_per_second_upper_estimate": 0.004031684371584596
+      "required_token_positions_per_second": 9625.780946371064,
+      "required_pages_per_second_upper_estimate": 0.00403011347308189
     },
     {
       "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": 19259.06596178486,
-      "required_pages_per_second_upper_estimate": 0.004011155544115866
+      "required_token_positions_per_second": 19251.561892742127,
+      "required_pages_per_second_upper_estimate": 0.004009592644429873
     },
     {
       "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": 28888.598942677294,
-      "required_pages_per_second_upper_estimate": 0.0017318273962769895
+      "required_token_positions_per_second": 28877.34283911319,
+      "required_pages_per_second_upper_estimate": 0.0017311526100553965
     },
     {
       "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": 28888.598942677294,
-      "required_pages_per_second_upper_estimate": 0.00224596150027532
+      "required_token_positions_per_second": 28877.34283911319,
+      "required_pages_per_second_upper_estimate": 0.002245086387733578
     },
     {
       "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": 28888.598942677294,
-      "required_pages_per_second_upper_estimate": 0.37055192907578527
+      "required_token_positions_per_second": 28877.34283911319,
+      "required_pages_per_second_upper_estimate": 0.3704075478651273
     }
   ],
   "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": 0.31447493699948836,
-    "producer_p95_seconds": 0.35096166899893433,
-    "gpu_update_seconds": 0.9570557594316369,
-    "producer_batches_per_second": 3.1799036500057456,
-    "gpu_batches_per_second": 1.0448712001836404,
-    "producer_headroom": 3.043345102675683,
+    "producer_batch_seconds": 0.2759469797343854,
+    "producer_p95_seconds": 0.34647825099818874,
+    "gpu_update_seconds": 0.9574288103319502,
+    "producer_batches_per_second": 3.6238845627611385,
+    "gpu_batches_per_second": 1.0444640783822767,
+    "producer_headroom": 3.469611485704716,
     "sustainable": true,
-    "buffer_seconds": 60.29451284419312,
+    "buffer_seconds": 60.31801505091286,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 493283.2379553452,
-    "upload_bytes_per_second": 5878939.775890135,
+    "generation_bytes_per_second": 474669.46285059606,
+    "upload_bytes_per_second": 5991469.253135811,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.84709981913821,
+    "estimated_full_shards_per_hour": 101.85301698816691,
     "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": 0.9811340400010522,
-      "bytes": 2572134,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.036350621998281,
       "bytes": 6380141,
@@ -562,6 +556,12 @@
       "bytes": 11848768,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1809732929978054,
+      "bytes": 12035253,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 517100, "loss": 0.6098175644874573, "lr": 0.0003, "grad_norm": 0.1698224621202967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75685702108, "tokens_per_second": 151532.1575203471, "data_wait_s": 1.554000482428819e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517200, "loss": 0.5774507522583008, "lr": 0.0003, "grad_norm": 0.15616675411343853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75700266071, "tokens_per_second": 151283.52760896494, "data_wait_s": 1.5423000149894506e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517300, "loss": 0.5869436860084534, "lr": 0.0003, "grad_norm": 0.158736958374673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75714707979, "tokens_per_second": 150045.42190867095, "data_wait_s": 2.1870997443329543e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 517400, "loss": 0.6182101368904114, "lr": 0.0003, "grad_norm": 0.17699378061598395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75729315612, "tokens_per_second": 152490.06762253493, "data_wait_s": 1.4882003597449511e-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:f5a189595a6e488b11dfa2926c20c43415e8bf97c76e280c353272cd44555171
-size 688977
+oid sha256:972eafe61e70898f36bfd903bc2e8f626126c098b04f58de58637671412a7ad4
+size 691095