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": 77459.36985729232,
-      "required_pages_per_second_upper_estimate": 0.005358818937227582
+      "required_token_positions_per_second": 77360.76148420978,
+      "required_pages_per_second_upper_estimate": 0.005351996981174784
     },
     {
       "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": 9682.42123216154,
-      "required_pages_per_second_upper_estimate": 0.004053827577958696
+      "required_token_positions_per_second": 9670.095185526223,
+      "required_pages_per_second_upper_estimate": 0.004048666919629612
     },
     {
       "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": 19364.84246432308,
-      "required_pages_per_second_upper_estimate": 0.004033186000080618
+      "required_token_positions_per_second": 19340.190371052446,
+      "required_pages_per_second_upper_estimate": 0.004028051619171764
     },
     {
       "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": 29047.263696484617,
-      "required_pages_per_second_upper_estimate": 0.0017413391060006873
+      "required_token_positions_per_second": 29010.28555657867,
+      "required_pages_per_second_upper_estimate": 0.0017391223229756762
     },
     {
       "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": 29047.263696484617,
-      "required_pages_per_second_upper_estimate": 0.0022582969869913427
+      "required_token_positions_per_second": 29010.28555657867,
+      "required_pages_per_second_upper_estimate": 0.0022554220992633037
     },
     {
       "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": 29047.263696484617,
-      "required_pages_per_second_upper_estimate": 0.372587110176686
+      "required_token_positions_per_second": 29010.28555657867,
+      "required_pages_per_second_upper_estimate": 0.3721127943020058
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5366441886093725,
-    "producer_p95_seconds": 0.2055995509999775,
-    "gpu_update_seconds": 0.9518280375354612,
-    "producer_batches_per_second": 1.8634320863351561,
-    "gpu_batches_per_second": 1.0506099427258615,
-    "producer_headroom": 1.7736669058170018,
+    "producer_batch_seconds": 0.10529636157029643,
+    "producer_p95_seconds": 0.1874482570001419,
+    "gpu_update_seconds": 0.9530412910303206,
+    "producer_batches_per_second": 9.49700431322496,
+    "gpu_batches_per_second": 1.0492724810683836,
+    "producer_headroom": 9.051037251596439,
     "sustainable": true,
-    "buffer_seconds": 60.916994402269516,
+    "buffer_seconds": 60.0416013349102,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 224663.9555499879,
-    "upload_bytes_per_second": 4818923.113755712,
+    "generation_bytes_per_second": 224457.541035282,
+    "upload_bytes_per_second": 4895894.279232627,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 48.207654951808244,
+    "estimated_full_shards_per_hour": 48.16336320203633,
     "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": 1.9814531229999375,
-      "bytes": 6026004,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.2945359899999858,
       "bytes": 2043786,
@@ -562,6 +556,12 @@
       "bytes": 12136915,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.7720820229988021,
+      "bytes": 11800483,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -102,3 +102,4 @@
 {"step": 852700, "loss": 0.5552854537963867, "lr": 3e-05, "grad_norm": 0.3149392184376773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124811340335, "tokens_per_second": 149904.9143558171, "data_wait_s": 1.1418000212870538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852800, "loss": 0.5362980365753174, "lr": 3e-05, "grad_norm": 0.3573354062209486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124826055684, "tokens_per_second": 142715.7767014391, "data_wait_s": 1.5141000403673388e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 852900, "loss": 0.5480002164840698, "lr": 3e-05, "grad_norm": 0.31473184307219726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124840690045, "tokens_per_second": 152391.08213359266, "data_wait_s": 1.203700048790779e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 853000, "loss": 0.5508236885070801, "lr": 3e-05, "grad_norm": 0.2984796562300817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124855451709, "tokens_per_second": 154012.14148950728, "data_wait_s": 1.1611000445554964e-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 sha256:55cf3c56016153f074f7839c38bd23baf0f61e54be29d0c27faf93f419153337
-size 222003
+oid sha256:e8159e4d6018e6ac769ae09ad7b2f2f74c3e3e41ad421b910dcaa9d5eca912fb
+size 224121